Jax_federated_learning.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Federated learning example with Weight Erosion scheme\n",
    "Based on an earlier version of [Sai Praneeth Karimireddy's Jax-based FL simulation framework](https://colab.research.google.com/drive/1tLFkhmWWbY5veOcqaeXkFbJh9M_4f0hQ) (GitHub: @saipraneet)\n",
    "\n",
    "\n",
    "## Document dependencies\n",
    "These cells needn't be executed. To document the used dependencies, execute these cells _after_ executing the import statements."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Requirement already satisfied: watermark in /home/grimberg/.local/lib/python3.7/site-packages (2.0.2)\n",
      "Requirement already satisfied: ipython in /opt/anaconda3/lib/python3.7/site-packages (from watermark) (7.8.0)\n",
      "Requirement already satisfied: decorator in /opt/anaconda3/lib/python3.7/site-packages (from ipython->watermark) (4.4.0)\n",
      "Requirement already satisfied: prompt-toolkit<2.1.0,>=2.0.0 in /opt/anaconda3/lib/python3.7/site-packages (from ipython->watermark) (2.0.10)\n",
      "Requirement already satisfied: pygments in /opt/anaconda3/lib/python3.7/site-packages (from ipython->watermark) (2.4.2)\n",
      "Requirement already satisfied: pexpect; sys_platform != \"win32\" in /opt/anaconda3/lib/python3.7/site-packages (from ipython->watermark) (4.7.0)\n",
      "Requirement already satisfied: backcall in /opt/anaconda3/lib/python3.7/site-packages (from ipython->watermark) (0.1.0)\n",
      "Requirement already satisfied: traitlets>=4.2 in /opt/anaconda3/lib/python3.7/site-packages (from ipython->watermark) (4.3.3)\n",
      "Requirement already satisfied: pickleshare in /opt/anaconda3/lib/python3.7/site-packages (from ipython->watermark) (0.7.5)\n",
      "Requirement already satisfied: jedi>=0.10 in /opt/anaconda3/lib/python3.7/site-packages (from ipython->watermark) (0.15.1)\n",
      "Requirement already satisfied: setuptools>=18.5 in /opt/anaconda3/lib/python3.7/site-packages (from ipython->watermark) (41.4.0)\n",
      "Requirement already satisfied: six>=1.9.0 in /opt/anaconda3/lib/python3.7/site-packages (from prompt-toolkit<2.1.0,>=2.0.0->ipython->watermark) (1.12.0)\n",
      "Requirement already satisfied: wcwidth in /opt/anaconda3/lib/python3.7/site-packages (from prompt-toolkit<2.1.0,>=2.0.0->ipython->watermark) (0.1.7)\n",
      "Requirement already satisfied: ptyprocess>=0.5 in /opt/anaconda3/lib/python3.7/site-packages (from pexpect; sys_platform != \"win32\"->ipython->watermark) (0.6.0)\n",
      "Requirement already satisfied: ipython-genutils in /opt/anaconda3/lib/python3.7/site-packages (from traitlets>=4.2->ipython->watermark) (0.2.0)\n",
      "Requirement already satisfied: parso>=0.5.0 in /opt/anaconda3/lib/python3.7/site-packages (from jedi>=0.10->ipython->watermark) (0.5.1)\n"
     ]
    }
   ],
   "source": [
    "!pip install --user watermark"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The watermark extension is already loaded. To reload it, use:\n",
      "  %reload_ext watermark\n",
      "\n",
      "CPython 3.7.4\n",
      "IPython 7.8.0\n",
      "\n",
      "numpy 1.18.5\n",
      "jax 0.1.70\n",
      "haiku 0.0.2\n",
      "tensorflow 2.1.0\n"
     ]
    }
   ],
   "source": [
    "%load_ext watermark\n",
    "print('')\n",
    "%watermark -v\n",
    "print('')\n",
    "%watermark -p numpy,jax,haiku,tensorflow"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Install dependencies"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Collecting jax\n",
      "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/da/91/81f4de19a0c305c1e848a0fadbb5ceb0f77298642a330746e4ef3d1f5d88/jax-0.1.75.tar.gz (426kB)\n",
      "\u001b[K     |████████████████████████████████| 430kB 2.8MB/s eta 0:00:01\n",
      "\u001b[?25hCollecting jaxlib\n",
      "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/15/04/eb2f52e5a4d9d33f3457a5077e3e3e5ad4dcab0449a6884829ed6ee16bf4/jaxlib-0.1.52-cp37-none-manylinux2010_x86_64.whl (33.1MB)\n",
      "\u001b[K     |████████████████████████████████| 33.1MB 43.8MB/s eta 0:00:01    |██████████▌                     | 10.8MB 4.8MB/s eta 0:00:05\n",
      "\u001b[?25hRequirement already satisfied, skipping upgrade: numpy>=1.12 in /home/grimberg/.local/lib/python3.7/site-packages (from jax) (1.18.5)\n",
      "Requirement already satisfied, skipping upgrade: absl-py in /opt/anaconda3/lib/python3.7/site-packages (from jax) (0.9.0)\n",
      "Requirement already satisfied, skipping upgrade: opt_einsum in /opt/anaconda3/lib/python3.7/site-packages (from jax) (3.2.1)\n",
      "Requirement already satisfied, skipping upgrade: scipy in /opt/anaconda3/lib/python3.7/site-packages (from jaxlib) (1.4.1)\n",
      "Requirement already satisfied, skipping upgrade: six in /opt/anaconda3/lib/python3.7/site-packages (from absl-py->jax) (1.12.0)\n",
      "Building wheels for collected packages: jax\n",
      "  Building wheel for jax (setup.py) ... \u001b[?25ldone\n",
      "\u001b[?25h  Created wheel for jax: filename=jax-0.1.75-cp37-none-any.whl size=492891 sha256=63351b17fdf69f98424159280715531ab03b9df2f51f49a375e631205f75193a\n",
      "  Stored in directory: /home/grimberg/.cache/pip/wheels/28/42/2b/b0fe61f9cbf74be61331602e1bdad30742dc44393e034df8f2\n",
      "Successfully built jax\n",
      "Installing collected packages: jax, jaxlib\n",
      "  Found existing installation: jax 0.1.70\n",
      "    Uninstalling jax-0.1.70:\n",
      "      Successfully uninstalled jax-0.1.70\n",
      "  Found existing installation: jaxlib 0.1.47\n",
      "    Uninstalling jaxlib-0.1.47:\n",
      "      Successfully uninstalled jaxlib-0.1.47\n",
      "Successfully installed jax-0.1.75 jaxlib-0.1.52\n"
     ]
    }
   ],
   "source": [
    "!pip install --upgrade --user jax jaxlib"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Collecting git+https://github.com/deepmind/dm-haiku\n",
      "  Cloning https://github.com/deepmind/dm-haiku to /tmp/pip-req-build-9v9xvrlo\n",
      "  Running command git clone -q https://github.com/deepmind/dm-haiku /tmp/pip-req-build-9v9xvrlo\n",
      "Requirement already satisfied, skipping upgrade: absl-py>=0.7.1 in /opt/anaconda3/lib/python3.7/site-packages (from dm-haiku==0.0.2) (0.9.0)\n",
      "Requirement already satisfied, skipping upgrade: numpy>=1.18.0 in /home/grimberg/.local/lib/python3.7/site-packages (from dm-haiku==0.0.2) (1.18.5)\n",
      "Requirement already satisfied, skipping upgrade: six in /opt/anaconda3/lib/python3.7/site-packages (from absl-py>=0.7.1->dm-haiku==0.0.2) (1.12.0)\n",
      "Building wheels for collected packages: dm-haiku\n",
      "  Building wheel for dm-haiku (setup.py) ... \u001b[?25ldone\n",
      "\u001b[?25h  Created wheel for dm-haiku: filename=dm_haiku-0.0.2-cp37-none-any.whl size=243543 sha256=1c6183ef50b0b1dab135939496bf6b2b80d1b2a1988851d076363f0e045417fc\n",
      "  Stored in directory: /tmp/pip-ephem-wheel-cache-444x8s9a/wheels/97/0f/e9/17f34e377f8d4060fa88a7e82bee5d8afbf7972384768a5499\n",
      "Successfully built dm-haiku\n",
      "Installing collected packages: dm-haiku\n",
      "  Found existing installation: dm-haiku 0.0.1\n",
      "    Uninstalling dm-haiku-0.0.1:\n",
      "      Successfully uninstalled dm-haiku-0.0.1\n",
      "Successfully installed dm-haiku-0.0.2\n"
     ]
    }
   ],
   "source": [
    "!pip install --upgrade --user git+https://github.com/deepmind/dm-haiku"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Collecting numpy\n",
      "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/50/8f/29d5688614f9bba59931683d5d353d738d4a3007833219ee19c455732753/numpy-1.19.1-cp37-cp37m-manylinux2010_x86_64.whl (14.5MB)\n",
      "\u001b[K     |████████████████████████████████| 14.5MB 3.3MB/s eta 0:00:01\n",
      "\u001b[?25hInstalling collected packages: numpy\n",
      "  Found existing installation: numpy 1.18.5\n",
      "    Uninstalling numpy-1.18.5:\n",
      "      Successfully uninstalled numpy-1.18.5\n",
      "\u001b[33m  WARNING: The scripts f2py, f2py3 and f2py3.7 are installed in '/home/grimberg/.local/bin' which is not on PATH.\n",
      "  Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location.\u001b[0m\n",
      "Successfully installed numpy-1.19.1\n"
     ]
    }
   ],
   "source": [
    "!pip install --upgrade --user numpy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "#!pip install --user setGPU"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Import packages (etc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "yRDenD5M0rDY"
   },
   "outputs": [],
   "source": [
    "from typing import Any, Generator, Tuple, Mapping, Sequence, Optional, Callable, Union, NamedTuple\n",
    "from collections import namedtuple\n",
    "import functools, inspect, time, random\n",
    "\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "from absl import app\n",
    "import haiku as hk\n",
    "import jax\n",
    "from jax.experimental import optix\n",
    "from jax.tree_util import tree_multimap, tree_map, tree_reduce, tree_leaves\n",
    "import jax.numpy as jnp\n",
    "from jax.lax import fori_loop\n",
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "yRDenD5M0rDY"
   },
   "outputs": [],
   "source": [
    "from module.custom_types import *\n",
    "import module.utils as utils"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "#import setGPU"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Misc type-related stuff"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "S-RWB9G48uJA"
   },
   "outputs": [],
   "source": [
    "# extracts messages from a list of client outputs.\n",
    "# @jax.partial(jax.jit, static_argnums=[1])  # fix extractor.\n",
    "def extract_from_cout(\n",
    "    couts: Sequence[ClientOutput],\n",
    "    extractor: Callable[[ClientOutput], Any]\n",
    "    ) -> Sequence[Any]:\n",
    "#   print(\"compiling: {}\".format(inspect.currentframe().f_code.co_name))\n",
    "  # TODO: use pytrees.transpose?\n",
    "  # msgs = fori_loop(0, len(couts),\n",
    "  #                    lambda i, msgs: msgs.append(extractor(couts[i])), [])\n",
    "  msgs = [extractor(cout) for cout in couts]\n",
    "  return msgs\n",
    "\n",
    "def init_client_states(client_data: Sequence[tf.data.Dataset]) -> Sequence[ClientState]:\n",
    "  return [ClientState() for i in range(len(client_data))]\n",
    "\n",
    "def make_jax_batch(batch: Mapping[str, tf.Tensor]) -> Batch:\n",
    "    \"\"\"Transform a Mapping[str, tf.Tensor] (tensorflow batch)\n",
    "    into a Mapping[str, jnp.array] (jax-compatible batch).\"\"\"\n",
    "    jax_batch = {}\n",
    "    for key in batch:\n",
    "        jax_batch[key] = jnp.array(batch[key].numpy(), dtype=np.float32)\n",
    "    return jax_batch\n",
    "\n",
    "def make_jax_dataset(ds: tf.data.Dataset) -> Sequence[Batch]:\n",
    "    jax_ds = []\n",
    "    for batch in ds:\n",
    "        jax_ds.append(make_jax_batch(batch))\n",
    "    return jax_ds"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Batch the data sets and create the test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "def make_dss(clients: Sequence[Tuple[np.ndarray, np.ndarray]],\n",
    "             hyperparams: ServerHyperParams,\n",
    "             requesting_client: int = 0,\n",
    "             test_ratio: float = 0.5\n",
    ") -> Tuple[DatasetSequence, tf.data.Dataset]:\n",
    "    \"\"\"Make batched `tf.data.Dataset`s based on the (X,y) tuples representing\n",
    "    each client's data. \n",
    "    \n",
    "    requesting_client: Index of the age group for which a test set is created.\n",
    "    The training set for the requesting client is placed at index 0 in the returned\n",
    "    DatasetSequence.\"\"\"\n",
    "    if requesting_client < 0 or requesting_client >= len(clients):\n",
    "        raise ValueError(\"Unexpected 'requesting_client' argument: {}\"\n",
    "                         \"\\nExpecting a non-negative int lower than \"\n",
    "                         \"{}.\".format(requesting_client, len(clients)))\n",
    "    \n",
    "    dss, ds_test = [None], None\n",
    "    for i, (X,y) in enumerate(clients):\n",
    "        if i == requesting_client:\n",
    "            test_size = 1 + int(y.shape[0] * test_ratio)\n",
    "            ds_test = utils.make_ds(X[:test_size], y[:test_size], hyperparams.batch_size)\n",
    "            dss[0] = utils.make_ds(X[test_size:], y[test_size:], hyperparams.batch_size)\n",
    "        else:\n",
    "            dss.append( utils.make_ds(X, y, hyperparams.batch_size) )\n",
    "    \n",
    "    assert (ds_test is not None)\n",
    "    \n",
    "    return dss, ds_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hyperparams = ServerHyperParams(\n",
    "#     num_rounds = 30,\n",
    "#     max_batches_per_round = 3,\n",
    "#     max_epochs_per_round = 1,\n",
    "#     batch_size = 10,\n",
    "#     seed = 420\n",
    "# )\n",
    "# dss, ds_test = make_dss(split_by_age(*load_2017annie_predict_EVD()), hyperparams)    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# for batch in ds_test:\n",
    "#     print(batch)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Model, loss function, training, etc."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "GbTmc5DwjoON"
   },
   "outputs": [],
   "source": [
    "# Neural network model.\n",
    "def net_fn(batch: Batch) -> jnp.ndarray:\n",
    "  \"\"\"Logistic regression for 2-class classification.\"\"\"\n",
    "  x = batch[\"features\"]\n",
    "  logistic_reg = hk.Sequential([\n",
    "      hk.Flatten(),\n",
    "      hk.Linear(2, with_bias=True), jax.nn.log_softmax\n",
    "  ])\n",
    "  return logistic_reg(x)\n",
    "net: hk.Transformed = hk.transform(net_fn)\n",
    "\n",
    "# Initialize neural network parameters \n",
    "def init(rng: jax.random.PRNGKey, batch: Batch) -> hk.Params:\n",
    "  return net.init(rng, batch)\n",
    "\n",
    "# get predictions from model.\n",
    "def forward(params: hk.Params, batch: Batch):\n",
    "  return jax.jit(net.apply)(params,\n",
    "                            None,   # apply requires a rng argument that is not needed here by net_fn.\n",
    "                            batch)\n",
    "\n",
    "\n",
    "# Training loss (cross-entropy).\n",
    "@jax.jit\n",
    "def loss(params: hk.Params, batch: Batch) -> jnp.ndarray:\n",
    "  \"\"\"Compute the loss of the network, including L2.\"\"\"\n",
    "  print(\"compiling: {}\".format(inspect.currentframe().f_code.co_name))\n",
    "  preds = forward(params, batch)\n",
    "  labels = hk.one_hot(batch[\"label\"], 2)\n",
    "  # TODO: Put weight decay into optimizer\n",
    "  l2_loss = 0.5 * sum(jnp.sum(jnp.square(p)) for p in jax.tree_leaves(params))\n",
    "  softmax_xent = -jnp.mean(labels * preds, dtype=np.float32)\n",
    "  return softmax_xent+ 1e-3 * l2_loss\n",
    "\n",
    "# Evaluation metric (classification accuracy).\n",
    "@jax.jit\n",
    "def accuracy(params: hk.Params, batch: Batch) -> jnp.ndarray:\n",
    "  print(\"compiling: {}\".format(inspect.currentframe().f_code.co_name))\n",
    "  preds = forward(params, batch)\n",
    "  pred_class = jnp.argmax(preds, axis=-1)\n",
    "  print(pred_class)\n",
    "  return jnp.mean(pred_class == batch[\"label\"], dtype=np.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "elJQ6h8MVeH6"
   },
   "outputs": [],
   "source": [
    "# one local update step.\n",
    "@jax.partial(jax.jit, static_argnums=[2,4])  # fix loss function and optimizer.\n",
    "def run_one_step(\n",
    "    params: hk.Params,\n",
    "    batch: Batch,\n",
    "    client_opt: optix.InitUpdate,\n",
    "    opt_state: OptState,\n",
    "    loss: LossFunction\n",
    "    ) -> Tuple[hk.Params, OptState]:\n",
    "  \"\"\"Learning rule (stochastic gradient descent).\"\"\"\n",
    "  print(\"compiling: {}\".format(inspect.currentframe().f_code.co_name))\n",
    "  grads = jax.grad(loss)(params, batch)\n",
    "  updates, opt_state = client_opt.update(grads, opt_state)\n",
    "  new_params = optix.apply_updates(params, updates)\n",
    "  return new_params, opt_state\n",
    "\n",
    "# perform client updates.\n",
    "def client_updater(\n",
    "    msg: ClientMessage,\n",
    "    ds: Sequence[Batch],\n",
    "    client_opt: optix.InitUpdate,\n",
    "    loss: LossFunction\n",
    "    ) -> ClientOutput:\n",
    "  opt_state = client_opt.init(*msg.opt_init_input)\n",
    "  # iterate through data making updates.\n",
    "  new_params = msg.params\n",
    "  train_loss, train_acc = [], []\n",
    "  for minibatch in ds:\n",
    "    new_params, opt_state = run_one_step(new_params,\n",
    "                                         minibatch,\n",
    "                                         client_opt, \n",
    "                                         opt_state, \n",
    "                                         loss)\n",
    "    train_loss.append(loss(msg.params, minibatch))\n",
    "    train_acc.append(accuracy(msg.params, minibatch))\n",
    "  # compute and return the change in parameters.\n",
    "  params_update = tree_multimap(lambda x, y: x - y, new_params, msg.params)\n",
    "  \n",
    "  # TODO: replace with an function which constructs message  \n",
    "  msg_to_server = ServerMessage(\n",
    "      aggregator_input=params_update,\n",
    "      stateupdater_input=None\n",
    "  )\n",
    "\n",
    "  diagnostic_msg = DiagnosticsMessage(\n",
    "      train_loss=np.mean(train_loss),\n",
    "      train_acc=np.mean(train_acc),\n",
    "      test_loss=0,\n",
    "      test_acc=1,\n",
    "      weight=len(train_loss)\n",
    "  )\n",
    "  return msg_to_server, diagnostic_msg"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Make clients stateful (remember which batch is up next)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Absolutely not suitable for jit-compilation.\n",
    "def make_client_data(\n",
    "    ds: tf.data.Dataset,\n",
    "    client_state: ClientState,\n",
    "    hyperparams: ServerHyperParams\n",
    ") -> Tuple[Sequence[Batch], ClientState]:\n",
    "  \"\"\"Picks out and assembles the batches used in the next round\n",
    "  for one client, updating the client state along the way.\n",
    "  \"\"\"\n",
    "  batches_before, batches_after = [], []\n",
    "  \n",
    "  # Initially skip all the way to client_state.next_idx:\n",
    "  for i, batch in enumerate(ds):\n",
    "    if i < client_state.next_idx:\n",
    "      batches_before.append(\n",
    "        make_jax_batch(batch)\n",
    "      )\n",
    "    # Return once num_batches batches have been selected:\n",
    "    elif len(batches_after) >= hyperparams.max_batches_per_round:\n",
    "      client_state = client_state._replace(next_idx = i)\n",
    "      return batches_after, client_state\n",
    "    # Transform the tf batch to a Batch (Mapping[str, jnp.array])\n",
    "    else:\n",
    "      batches_after.append(\n",
    "        make_jax_batch(batch)\n",
    "      )\n",
    "  \n",
    "  # Increment the client's epoch count once, because we've reached\n",
    "  # the end of the dataset:\n",
    "  client_state = client_state._replace(\n",
    "                            next_idx = 0,\n",
    "                            epoch_count = client_state.epoch_count + 1\n",
    "                            )\n",
    "  \n",
    "  # Traverse the dataset up to max_epochs-1 more times, until num_batches\n",
    "  # batches are in batches_after\n",
    "  for epoch in range(hyperparams.max_epochs_per_round - 1):\n",
    "    \n",
    "    for i, batch in enumerate(ds):\n",
    "      # Return once num_batches batches have been selected:\n",
    "      if len(batches_after) >= hyperparams.max_batches_per_round:\n",
    "        client_state = client_state._replace(next_idx = i)\n",
    "        return batches_after, client_state\n",
    "      # Transform the tf batch to a Batch (Mapping[str, jnp.array])\n",
    "      batches_after.append(\n",
    "        make_jax_batch(batch)\n",
    "      )\n",
    "    \n",
    "    # Increment the client's epoch_count\n",
    "    client_state = client_state._replace(\n",
    "                            next_idx = 0,\n",
    "                            epoch_count = client_state.epoch_count + 1\n",
    "                            )\n",
    "  \n",
    "  # If num_batches has not yet been reached,\n",
    "  # add (some or all of) the batches_before at the end:\n",
    "  num_to_add = hyperparams.max_batches_per_round - len(batches_after)\n",
    "  batches_to_add = batches_before[:num_to_add]\n",
    "  client_state = client_state._replace(next_idx = len(batches_to_add))\n",
    "  batches_after += batches_to_add\n",
    "  return batches_after, client_state\n",
    "\n",
    "def make_all_client_data(\n",
    "    client_data: Sequence[tf.data.Dataset],\n",
    "    client_states: Sequence[ClientState],\n",
    "    hyperparams: ServerHyperParams\n",
    ") -> Tuple[Sequence[Sequence[Batch]], Sequence[ClientState]]:\n",
    "  \"\"\"Picks out and assembles the batches used in the next round\n",
    "  for all clients, updating their states along the way.\n",
    "  \"\"\"\n",
    "  if not (len(client_states) == len(client_data)):\n",
    "    raise ValueError(\n",
    "        \"Number of datasets ({}) does not fit number of client \"\n",
    "        \"states ({})!\".format(len(client_data), len(client_states))\n",
    "        )\n",
    "  dss_jax, client_states_after = [], []\n",
    "  for ds, client_state_before in zip(client_data, client_states):\n",
    "    batches_jax, client_state_after = make_client_data(ds, client_state_before, hyperparams)\n",
    "    dss_jax.append(batches_jax)\n",
    "    client_states_after.append(client_state_after)\n",
    "  return dss_jax, client_states_after"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Aggregators"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "zvRbtDM1wS_s"
   },
   "outputs": [],
   "source": [
    "# aggregate client updates.\n",
    "# TODO: make the aggregator stateful.\n",
    "@jax.jit\n",
    "def average_params(params_list: Sequence[hk.Params],\n",
    "                   client_states: Sequence[ClientState],\n",
    "                   hyperparams: AverageAggregatorHyperParams\n",
    "                  ) -> hk.Params: \n",
    "  print(\"compiling: {}\".format(inspect.currentframe().f_code.co_name))\n",
    "  num_params = len(params_list)\n",
    "  params_sum = functools.reduce(\n",
    "      lambda t1, t2: tree_multimap(sum, t1, t2), params_list)\n",
    "  params_avg = tree_map(lambda x: x/num_params, params_sum)\n",
    "  print('num_params:', num_params)\n",
    "#   print('params_list:', params_list)\n",
    "#   print('params_avg:',  params_avg)\n",
    "  return params_avg, client_states\n",
    "\n",
    "@jax.jit\n",
    "def normalize_array_l2(x: jnp.array) -> jnp.array:\n",
    "    print(\"compiling: {}\".format(inspect.currentframe().f_code.co_name))\n",
    "    return x / jnp.sqrt(jnp.dot(x, x))\n",
    "\n",
    "@jax.jit\n",
    "def normalize_array_l1(x: jnp.array) -> jnp.array:\n",
    "    print(\"compiling: {}\".format(inspect.currentframe().f_code.co_name))\n",
    "    return x / jnp.sum(jnp.abs(x))\n",
    "\n",
    "@jax.jit\n",
    "def tree_norm(tree: hk.Params) -> jnp.array:\n",
    "    print(\"compiling: {}\".format(inspect.currentframe().f_code.co_name))\n",
    "    return jnp.sqrt(\n",
    "               jnp.sum(\n",
    "                   tree_leaves(\n",
    "                       tree_map(\n",
    "                           lambda x: jnp.sum(x*x),\n",
    "                           tree\n",
    "                       )\n",
    "                   )\n",
    "               )\n",
    "           )\n",
    "\n",
    "# @jax.jit\n",
    "def calc_rel_distances_of_others(reference_update: hk.Params,\n",
    "                                 other_updates: Sequence[hk.Params]\n",
    "                                ) -> Sequence[Any]:\n",
    "#   print(\"compiling: {}\".format(inspect.currentframe().f_code.co_name))\n",
    "  # Compare the distances to the norm of the reference update.\n",
    "  reference_norm = tree_norm(reference_update)\n",
    "  return [ tree_norm(tree_multimap(lambda x, y: x-y, update, reference_update))\n",
    "           / reference_norm \n",
    "           for update in other_updates ]\n",
    "\n",
    "# @jax.partial\n",
    "def sum_args(*args):\n",
    "#   print(\"compiling: {} with {} arguments\".format(inspect.currentframe().f_code.co_name, len(args)))\n",
    "  return sum(args)\n",
    "\n",
    "# @jax.jit\n",
    "def similarity_aggregator(updates_list: Sequence[hk.Params],\n",
    "                          client_states: Sequence[ClientState],\n",
    "                          hyperparams: SimilarityAggregatorHyperParams\n",
    "                         ) -> Tuple[hk.Params, Sequence[ClientState]]:\n",
    "#     print(\"compiling: {}\".format(inspect.currentframe().f_code.co_name))\n",
    "    assert len(updates_list) == len(client_states)\n",
    "    reference_update = updates_list[0]\n",
    "    other_updates = updates_list[1:]\n",
    "    rel_distances_of_others = calc_rel_distances_of_others(reference_update,\n",
    "                                                           other_updates)\n",
    "    print(\"rel_distances_of_others:\", rel_distances_of_others)\n",
    "    \n",
    "    # Each dataset starts with a similarity of 1, which can only decrease over time.\n",
    "    # At every round, each dataset's similarity could decrease by something like:\n",
    "    # some_factor * (1+client_state.epoch_count) * rel_distance\n",
    "    for client_id, rel_distance in enumerate(rel_distances_of_others, start=1):\n",
    "        c = client_states[client_id]\n",
    "        print(c.similarity)\n",
    "        decrement = (  rel_distance\n",
    "                     * hyperparams.distance_penalty_factor\n",
    "                     * (1 + c.epoch_count*hyperparams.size_penalty_factor)\n",
    "                    )\n",
    "        similarity = jnp.max([0, c.similarity - decrement], dtype=np.float32).astype(np.float32)\n",
    "        client_states[client_id] = c._replace(similarity=similarity)\n",
    "    \n",
    "    print([c.similarity for c in client_states])\n",
    "    weights = normalize_array_l1(jnp.array([c.similarity for c in client_states], dtype=np.float32))\n",
    "    weighted_updates = [tree_map(lambda x: w*x, update) for w, update in zip(weights, updates_list)]\n",
    "    aggregated_update = tree_multimap(sum_args, *weighted_updates)\n",
    "    return aggregated_update, client_states\n",
    "\n",
    "def init_average_aggregator():\n",
    "    return Aggregator(aggregator_function    = average_params,\n",
    "                      aggregator_hyperparams = AverageAggregatorHyperParams()\n",
    "                     )\n",
    "\n",
    "def init_similarity_aggregator(distance_penalty_factor: float = 0.01,\n",
    "                               size_penalty_factor    : float = 0.2\n",
    "                              ) -> Aggregator:\n",
    "    similarity_hyperparams = SimilarityAggregatorHyperParams(\n",
    "            distance_penalty_factor=distance_penalty_factor,\n",
    "            size_penalty_factor=size_penalty_factor\n",
    "    )\n",
    "    return Aggregator(aggregator_function    = similarity_aggregator,\n",
    "                      aggregator_hyperparams = similarity_hyperparams\n",
    "                     )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "# # initialize random generator, params, opt_state.\n",
    "# num_splits = 4\n",
    "# rng = jax.random.PRNGKey(hyperparams.seed)\n",
    "# rngs = jax.random.split(rng, num_splits+1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "# updates_list = [init(rngs[i], dss[0].as_numpy_iterator().next()) for i in range(num_splits+1)]\n",
    "# client_states = [ClientState() for i in range(num_splits+1)]\n",
    "# client_states = similarity_aggregator(updates_list, client_states, SimilarityAggregatorHyperParams(0.05))[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "# for i in range(3):\n",
    "#     reference_update = init(rngs[-1], dss[0].as_numpy_iterator().next())\n",
    "#     other_updates = [init(rngs[i], dss[0].as_numpy_iterator().next()) for i in range(num_splits)]\n",
    "#     print(calc_rel_distances_of_others(reference_update, other_updates))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    " DeviceArray(1.3672365, dtype=float32), DeviceArray(1.3019154, dtype=float32), DeviceArray(1.5341283, dtype=float32), DeviceArray(1.6079143, dtype=float32)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# for i in range(3):\n",
    "#     reference_update = init(rngs[0], dss[0].as_numpy_iterator().next())\n",
    "#     other_updates = [init(rngs[i+1], dss[0].as_numpy_iterator().next()) for i in range(num_splits)]\n",
    "#     print(calc_rel_distances_of_others(reference_update, other_updates))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# One round of training & communication"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "c9V22gJC2DIn"
   },
   "outputs": [],
   "source": [
    "# aggregate diagnositcs.\n",
    "def agg_diagnostics(\n",
    "    client_outputs: Sequence[ClientOutput]) -> DiagnosticsMessage:\n",
    "  d_msgs = extract_from_cout(client_outputs, lambda cout: cout[1])\n",
    "  # TODO: aggregate and report statistics\n",
    "  total_weights = np.sum([msg.weight for msg in d_msgs])\n",
    "\n",
    "  train_loss = np.sum([msg.train_loss for msg in d_msgs]) / total_weights\n",
    "  train_acc =  np.sum([msg.train_acc for msg in d_msgs])  / total_weights\n",
    "  return DiagnosticsMessage(train_loss=train_loss,\n",
    "                            train_acc = train_acc,\n",
    "                            test_loss=0,\n",
    "                            test_acc=1,\n",
    "                            weight=total_weights\n",
    "                           )\n",
    "\n",
    "\n",
    "# update server params.\n",
    "# @jax.partial(jax.jit, static_argnums=[3,5])  # fix optimizer and aggregator.\n",
    "def server_updater(\n",
    "    server_params: hk.Params,\n",
    "    client_outputs: Sequence[ClientOutput],\n",
    "    client_states: Sequence[ClientState],\n",
    "    server_opt: optix.InitUpdate,\n",
    "    opt_state: OptState,\n",
    "    aggregator: Aggregator\n",
    ") -> Tuple[hk.Params, Sequence[ClientState], OptState]:\n",
    "#   print(\"compiling: {}\".format(inspect.currentframe().f_code.co_name))\n",
    "  agg_inputs = extract_from_cout(client_outputs,\n",
    "                                 lambda cout: cout[0].aggregator_input)\n",
    "  agg_update, client_states_out = aggregator.aggregator_function(\n",
    "                                                agg_inputs,\n",
    "                                                client_states,\n",
    "                                                aggregator.aggregator_hyperparams\n",
    "                                                )\n",
    "  eff_grads = tree_map(lambda x: -1.0*x, agg_update)  # effective gradient.\n",
    "  updates, opt_state = server_opt.update(eff_grads, opt_state)\n",
    "  # TODO: allow opt_state to also be explicitly updated by state_updater\n",
    "  server_params = optix.apply_updates(server_params, updates)\n",
    "  return server_params, client_states_out, opt_state\n",
    "\n",
    "# one round of federated learning.\n",
    "def run_one_round(\n",
    "    server_params: hk.Params,\n",
    "    hyperparams: ServerHyperParams,\n",
    "    client_data: Sequence[tf.data.Dataset],\n",
    "    client_states: Sequence[ClientState],\n",
    "    client_opt: optix.InitUpdate,\n",
    "    server_opt: optix.InitUpdate,\n",
    "    opt_state: OptState,\n",
    "    aggregator: Aggregator,\n",
    "    loss: LossFunction,\n",
    "    rng: jax.random.PRNGKey\n",
    ") -> Tuple[hk.Params, Sequence[ClientState], OptState]:\n",
    "  \n",
    "  print(\"computing updates from active clients.\")\n",
    "  # TODO: replace with an function which constructs message\n",
    "  msg_to_clients = ClientMessage(\n",
    "      params=server_params, \n",
    "      opt_init_input=[server_params])\n",
    "  \n",
    "  dss, client_states_upd = make_all_client_data(client_data, client_states, hyperparams)\n",
    "  \n",
    "  client_outputs = list(map(lambda ds: client_updater(\n",
    "                          msg_to_clients,\n",
    "                          ds,\n",
    "                          client_opt,\n",
    "                          loss), dss))\n",
    "  \n",
    "  print(\"aggregating client updates.\")\n",
    "  server_params, client_states_out, opt_state = server_updater(\n",
    "      server_params,\n",
    "      client_outputs,\n",
    "      client_states_upd,\n",
    "      server_opt,      \n",
    "      opt_state,      \n",
    "      aggregator)\n",
    "  \n",
    "  # TODO: aggregate and incporate new diagnostics. needs state!\n",
    "  diagnostics = agg_diagnostics(client_outputs)\n",
    "  \n",
    "  return server_params, client_states_out, opt_state, diagnostics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 383
    },
    "colab_type": "code",
    "id": "BuZcD4dEdzF3",
    "outputId": "08edffde-2229-4389-e8b7-bcd7c00320e2"
   },
   "outputs": [],
   "source": [
    "# # Testing one round of federated averaging.\n",
    "# batch_size=10\n",
    "# dss, ds_test = split_by_age(*load_2017annie_predict_EVD(), batch_size)\n",
    "# client_states = init_client_states(dss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 383
    },
    "colab_type": "code",
    "id": "BuZcD4dEdzF3",
    "outputId": "08edffde-2229-4389-e8b7-bcd7c00320e2"
   },
   "outputs": [],
   "source": [
    "# params = init(jax.random.PRNGKey(42), dss[0].as_numpy_iterator().next())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 383
    },
    "colab_type": "code",
    "id": "BuZcD4dEdzF3",
    "outputId": "08edffde-2229-4389-e8b7-bcd7c00320e2"
   },
   "outputs": [],
   "source": [
    "# client_opt = optix.sgd(0.1)\n",
    "# server_opt = optix.sgd(1.0)\n",
    "# opt_state = server_opt.init(params)\n",
    "# rng = jax.random.PRNGKey(0)\n",
    "# hyperparams = ServerHyperParams(\n",
    "#     num_rounds = 12,\n",
    "#     max_batches_per_round = 5000,\n",
    "#     max_epochs_per_round = 1,\n",
    "#     batch_size = 10,\n",
    "#     seed = 7\n",
    "# )\n",
    "\n",
    "# run_one_round(\n",
    "#     params,\n",
    "#     hyperparams,\n",
    "#     dss,\n",
    "#     client_states,\n",
    "#     client_opt,\n",
    "#     server_opt,\n",
    "#     opt_state,\n",
    "#     average_params,\n",
    "#     loss,\n",
    "#     rng\n",
    "# );"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Test metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "def eval_test_accuracy(params: hk.Params, ds: Sequence[Batch]) -> np.float64:\n",
    "    accs = []\n",
    "    for batch in ds:\n",
    "        accs.append(accuracy(params, batch))\n",
    "    return np.mean(accs)\n",
    "\n",
    "def one_hot_encode(labels, n_classes):\n",
    "    \"\"\"\n",
    "    One hot encode a list of sample labels. Return a one-hot encoded vector for each label.\n",
    "    : labels: List of sample labels\n",
    "    : return: Numpy array of one-hot encoded labels\n",
    "     \"\"\"\n",
    "    targets = np.array(labels, dtype=np.int8).reshape(-1)\n",
    "    return np.eye(n_classes)[targets]\n",
    "\n",
    "# Baseline classification accuracy (prevalence of most prevalent label)\n",
    "def eval_baseline_test_accuracy(ds: tf.data.Dataset) -> np.float64:\n",
    "    labels = one_hot_encode(np.concatenate([batch[\"label\"] for batch in ds], axis=0), 2)\n",
    "    return np.max(labels.mean(axis=0))\n",
    "\n",
    "\n",
    "@jax.jit\n",
    "def baseline_loss(batch: Batch, mean_label) -> jnp.ndarray:\n",
    "  \"\"\"Compute the loss of the baseline model.\"\"\"\n",
    "  print(\"compiling: {}\".format(inspect.currentframe().f_code.co_name))\n",
    "  preds = jnp.stack((jnp.full_like(batch[\"label\"], 1.-mean_label), jnp.full_like(batch[\"label\"], mean_label)), axis=1)\n",
    "  labels = hk.one_hot(batch[\"label\"], 2)\n",
    "  # TODO: Put weight decay into optimizer\n",
    "  softmax_xent = -jnp.mean(labels * preds, dtype=np.float32)\n",
    "  return softmax_xent\n",
    "\n",
    "def eval_baseline_loss(ds: tf.data.Dataset) -> np.float64:\n",
    "    bsizes = [batch['label'].shape[0] for batch in ds]\n",
    "    mean_label = sum(np.mean(batch['label'])*bsizes[i] for i, batch in enumerate(ds)) / np.sum(bsizes)\n",
    "    loss = sum(baseline_loss(make_jax_batch(batch), mean_label) * bsizes[i] for i, batch in enumerate(ds)) / np.sum(bsizes)\n",
    "    return loss.item()\n",
    "\n",
    "def eval_test_loss(params: hk.Params, ds: Sequence[Batch]) -> np.float64:\n",
    "    losses = []\n",
    "    for batch in ds:\n",
    "        losses.append(loss(params, batch))\n",
    "    return np.mean(losses)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Put it all together"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "def update_diagnostics(diag_state: DiagnosticsState,\n",
    "                       diagnostics: DiagnosticsMessage,\n",
    "                       client_states: Sequence[ClientState],\n",
    "                       server_params: hk.Params,\n",
    "                       ds_test: Sequence[Batch]\n",
    "                      ) -> DiagnosticsState:\n",
    "    diag_state_upd = DiagnosticsState(\n",
    "                        train_loss_global     = diag_state.train_loss_global     + [diagnostics.train_loss],\n",
    "                        train_accuracy_global = diag_state.train_accuracy_global + [diagnostics.train_acc],\n",
    "                        test_loss             = diag_state.test_loss     + [eval_test_loss(    server_params, ds_test)],\n",
    "                        test_accuracy         = diag_state.test_accuracy + [eval_test_accuracy(server_params, ds_test)],\n",
    "                        similarities          = diag_state.similarities + [[float(c.similarity) for c in client_states]],\n",
    "                        epoch_counts          = diag_state.epoch_counts + [[int(c.epoch_count) for c in client_states]]\n",
    "                       )\n",
    "    return diag_state_upd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "MsHof54fApTp"
   },
   "outputs": [],
   "source": [
    "def federated_learning(\n",
    "    client_data: Sequence[tf.data.Dataset],\n",
    "     aggregator: Aggregator,\n",
    "        ds_test: Sequence[Batch],\n",
    "    hyperparams: ServerHyperParams,\n",
    "     client_opt: optix.InitUpdate,\n",
    "     server_opt: optix.InitUpdate,\n",
    "           loss: LossFunction,\n",
    "           init\n",
    "    ) -> hk.Params:\n",
    "  # initialize random generator, params, opt_state.\n",
    "  rng = jax.random.PRNGKey(hyperparams.seed)\n",
    "  # TODO: use synthethic data for init\n",
    "  server_params = init(rng, client_data[0].as_numpy_iterator().next())\n",
    "  client_states = init_client_states(client_data)\n",
    "  opt_state = server_opt.init(server_params)\n",
    "  rngs = jax.random.split(rng, hyperparams.num_rounds)\n",
    "  \n",
    "  # Initialize the diagnostics with the test metrics of the randomly initialized model.\n",
    "  # At each round, the training loss and accuracy are computed based on the\n",
    "  # server_params *before* the round, while the test diagnostics are computed\n",
    "  # from the server_params *after* the round. Therefore, we do not initialize the\n",
    "  # training metrics here.\n",
    "  diag_state = DiagnosticsState(\n",
    "          test_loss     = [    eval_test_loss(server_params, ds_test)],\n",
    "          test_accuracy = [eval_test_accuracy(server_params, ds_test)],\n",
    "          similarities  = [[1]*len(client_data)],\n",
    "          epoch_counts  = [[0]*len(client_data)]\n",
    "  )\n",
    "  for round_num, rng in enumerate(rngs):\n",
    "    print(\"\\nrunning round {}\".format(round_num))\n",
    "    server_params, \\\n",
    "    client_states, \\\n",
    "    opt_state,     \\\n",
    "    diagnostics,     = run_one_round(server_params, hyperparams, \n",
    "                                    client_data, client_states, client_opt, \n",
    "                                    server_opt, opt_state, \n",
    "                                    aggregator, loss, rng)\n",
    "    diag_state = update_diagnostics(diag_state, diagnostics, client_states, server_params, ds_test)\n",
    "\n",
    "  return server_params, diag_state"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Run a federated learning simulation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 530
    },
    "colab_type": "code",
    "id": "iT-dXeovX2Fd",
    "outputId": "57c040e3-f579-497a-bb96-d30bed2d2004",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CLIENT 0 baseline test accuracy: 0.544\n",
      "\n",
      "\n",
      " --------------- TRAINING WITH Global (lr=0.07) AGGREGATOR FOR REQUESTING CLIENT = 0 ---------------\n",
      "compiling: loss\n",
      "compiling: accuracy\n",
      "Traced<ShapedArray(int32[125]):JaxprTrace(level=-1/1)>\n",
      "\n",
      "running round 0\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "compiling: loss\n",
      "compiling: loss\n",
      "compiling: accuracy\n",
      "Traced<ShapedArray(int32[123]):JaxprTrace(level=-1/1)>\n",
      "compiling: run_one_step\n",
      "compiling: loss\n",
      "compiling: loss\n",
      "compiling: accuracy\n",
      "Traced<ShapedArray(int32[161]):JaxprTrace(level=-1/1)>\n",
      "aggregating client updates.\n",
      "compiling: average_params\n",
      "num_params: 4\n",
      "\n",
      "running round 1\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "compiling: loss\n",
      "compiling: loss\n",
      "compiling: accuracy\n",
      "Traced<ShapedArray(int32[102]):JaxprTrace(level=-1/1)>\n",
      "aggregating client updates.\n",
      "compiling: average_params\n",
      "num_params: 4\n",
      "\n",
      "running round 2\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "compiling: loss\n",
      "compiling: loss\n",
      "compiling: accuracy\n",
      "Traced<ShapedArray(int32[154]):JaxprTrace(level=-1/1)>\n",
      "aggregating client updates.\n",
      "\n",
      "running round 3\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 4\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 5\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 6\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 7\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 8\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 9\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 10\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 11\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 12\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 13\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 14\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 15\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 16\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 17\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 18\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 19\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 20\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 21\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 22\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 23\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 24\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 25\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 26\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 27\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 28\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 29\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "\n",
      " --------------- TRAINING WITH Weight Erosion (lr=3.0) AGGREGATOR FOR REQUESTING CLIENT = 0 ---------------\n",
      "\n",
      "running round 0\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "compiling: tree_norm\n",
      "rel_distances_of_others: [DeviceArray(0.81650364, dtype=float32), DeviceArray(0.91607934, dtype=float32), DeviceArray(1.3923907, dtype=float32)]\n",
      "1.0\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/grimberg/.local/lib/python3.7/site-packages/jax/numpy/lax_numpy.py:1621: FutureWarning: jax.numpy reductions won't accept lists and tuples in future versions, only scalars and ndarrays\n",
      "  warnings.warn(msg, category=FutureWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.0\n",
      "1.0\n",
      "[1.0, DeviceArray(0.99183494, dtype=float32), DeviceArray(0.9908392, dtype=float32), DeviceArray(0.9860761, dtype=float32)]\n",
      "compiling: normalize_array_l1\n",
      "\n",
      "running round 1\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.866929, dtype=float32), DeviceArray(0.85813516, dtype=float32), DeviceArray(1.560172, dtype=float32)]\n",
      "0.99183494\n",
      "0.9908392\n",
      "0.9860761\n",
      "[1.0, DeviceArray(0.9831656, dtype=float32), DeviceArray(0.9805416, dtype=float32), DeviceArray(0.96735406, dtype=float32)]\n",
      "\n",
      "running round 2\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.99027264, dtype=float32), DeviceArray(0.9650648, dtype=float32), DeviceArray(1.1827812, dtype=float32)]\n",
      "0.9831656\n",
      "0.9805416\n",
      "0.96735406\n",
      "[1.0, DeviceArray(0.97128236, dtype=float32), DeviceArray(0.9689608, dtype=float32), DeviceArray(0.9531607, dtype=float32)]\n",
      "\n",
      "running round 3\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.1772193, dtype=float32), DeviceArray(1.0301032, dtype=float32), DeviceArray(2.133797, dtype=float32)]\n",
      "0.97128236\n",
      "0.9689608\n",
      "0.9531607\n",
      "[1.0, DeviceArray(0.9571557, dtype=float32), DeviceArray(0.95453936, dtype=float32), DeviceArray(0.9232876, dtype=float32)]\n",
      "\n",
      "running round 4\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.3843757, dtype=float32), DeviceArray(1.340707, dtype=float32), DeviceArray(1.2269342, dtype=float32)]\n",
      "0.9571557\n",
      "0.95453936\n",
      "0.9232876\n",
      "[1.0, DeviceArray(0.9405432, dtype=float32), DeviceArray(0.93576944, dtype=float32), DeviceArray(0.90611047, dtype=float32)]\n",
      "\n",
      "running round 5\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.1461687, dtype=float32), DeviceArray(1.582524, dtype=float32), DeviceArray(2.3357627, dtype=float32)]\n",
      "0.9405432\n",
      "0.93576944\n",
      "0.90611047\n",
      "[1.0, DeviceArray(0.9244968, dtype=float32), DeviceArray(0.910449, dtype=float32), DeviceArray(0.8687383, dtype=float32)]\n",
      "\n",
      "running round 6\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.5837946, dtype=float32), DeviceArray(1.5264198, dtype=float32), DeviceArray(1.2595572, dtype=float32)]\n",
      "0.9244968\n",
      "0.910449\n",
      "0.8687383\n",
      "[1.0, DeviceArray(0.9023237, dtype=float32), DeviceArray(0.8860263, dtype=float32), DeviceArray(0.84858537, dtype=float32)]\n",
      "\n",
      "running round 7\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.6802949, dtype=float32), DeviceArray(1.6765314, dtype=float32), DeviceArray(2.166084, dtype=float32)]\n",
      "0.9023237\n",
      "0.8860263\n",
      "0.84858537\n",
      "[1.0, DeviceArray(0.8787996, dtype=float32), DeviceArray(0.8558488, dtype=float32), DeviceArray(0.8095959, dtype=float32)]\n",
      "\n",
      "running round 8\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.4094106, dtype=float32), DeviceArray(1.6809635, dtype=float32), DeviceArray(1.4293803, dtype=float32)]\n",
      "0.8787996\n",
      "0.8558488\n",
      "0.8095959\n",
      "[1.0, DeviceArray(0.85624903, dtype=float32), DeviceArray(0.82559144, dtype=float32), DeviceArray(0.78386706, dtype=float32)]\n",
      "\n",
      "running round 9\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.0392141, dtype=float32), DeviceArray(1.6130265, dtype=float32), DeviceArray(2.295228, dtype=float32)]\n",
      "0.85624903\n",
      "0.82559144\n",
      "0.78386706\n",
      "[1.0, DeviceArray(0.82362163, dtype=float32), DeviceArray(0.7933309, dtype=float32), DeviceArray(0.7379625, dtype=float32)]\n",
      "\n",
      "running round 10\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.9549187, dtype=float32), DeviceArray(1.7231077, dtype=float32), DeviceArray(1.4936125, dtype=float32)]\n",
      "0.82362163\n",
      "0.7933309\n",
      "0.7379625\n",
      "[1.0, DeviceArray(0.7923429, dtype=float32), DeviceArray(0.75886875, dtype=float32), DeviceArray(0.70809025, dtype=float32)]\n",
      "\n",
      "running round 11\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.3056493, dtype=float32), DeviceArray(1.6887197, dtype=float32), DeviceArray(2.0864477, dtype=float32)]\n",
      "0.7923429\n",
      "0.75886875\n",
      "0.70809025\n",
      "[1.0, DeviceArray(0.7688412, dtype=float32), DeviceArray(0.72171694, dtype=float32), DeviceArray(0.6621884, dtype=float32)]\n",
      "\n",
      "running round 12\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.998094, dtype=float32), DeviceArray(1.7395781, dtype=float32), DeviceArray(1.495138, dtype=float32)]\n",
      "0.7688412\n",
      "0.72171694\n",
      "0.6621884\n",
      "[1.0, DeviceArray(0.7328755, dtype=float32), DeviceArray(0.6834462, dtype=float32), DeviceArray(0.62929535, dtype=float32)]\n",
      "\n",
      "running round 13\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.8120308, dtype=float32), DeviceArray(1.6571689, dtype=float32), DeviceArray(1.9196699, dtype=float32)]\n",
      "0.7328755\n",
      "0.6834462\n",
      "0.62929535\n",
      "[1.0, DeviceArray(0.700259, dtype=float32), DeviceArray(0.6436742, dtype=float32), DeviceArray(0.5832233, dtype=float32)]\n",
      "\n",
      "running round 14\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.5528783, dtype=float32), DeviceArray(1.8259, dtype=float32), DeviceArray(1.6472427, dtype=float32)]\n",
      "0.700259\n",
      "0.6436742\n",
      "0.5832233\n",
      "[1.0, DeviceArray(0.66920143, dtype=float32), DeviceArray(0.59985256, dtype=float32), DeviceArray(0.5436894, dtype=float32)]\n",
      "\n",
      "running round 15\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.3453324, dtype=float32), DeviceArray(1.6808052, dtype=float32), DeviceArray(2.1800444, dtype=float32)]\n",
      "0.66920143\n",
      "0.59985256\n",
      "0.5436894\n",
      "[1.0, DeviceArray(0.6222948, dtype=float32), DeviceArray(0.5561516, dtype=float32), DeviceArray(0.48700827, dtype=float32)]\n",
      "\n",
      "running round 16\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.1229248, dtype=float32), DeviceArray(1.8576205, dtype=float32), DeviceArray(1.6985016, dtype=float32)]\n",
      "0.6222948\n",
      "0.5561516\n",
      "0.48700827\n",
      "[1.0, DeviceArray(0.5798363, dtype=float32), DeviceArray(0.5078535, dtype=float32), DeviceArray(0.44284722, dtype=float32)]\n",
      "\n",
      "running round 17\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.4242194, dtype=float32), DeviceArray(1.7113006, dtype=float32), DeviceArray(1.9766265, dtype=float32)]\n",
      "0.5798363\n",
      "0.5078535\n",
      "0.44284722\n",
      "[1.0, DeviceArray(0.54850346, dtype=float32), DeviceArray(0.4599371, dtype=float32), DeviceArray(0.3875017, dtype=float32)]\n",
      "\n",
      "running round 18\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.3795514, dtype=float32), DeviceArray(1.9491491, dtype=float32), DeviceArray(1.7531499, dtype=float32)]\n",
      "0.54850346\n",
      "0.4599371\n",
      "0.3875017\n",
      "[1.0, DeviceArray(0.49615332, dtype=float32), DeviceArray(0.40536094, dtype=float32), DeviceArray(0.3384135, dtype=float32)]\n",
      "\n",
      "running round 19\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.9394424, dtype=float32), DeviceArray(1.6875709, dtype=float32), DeviceArray(1.7412401, dtype=float32)]\n",
      "0.49615332\n",
      "0.40536094\n",
      "0.3384135\n",
      "[1.0, DeviceArray(0.4534856, dtype=float32), DeviceArray(0.35473382, dtype=float32), DeviceArray(0.2861763, dtype=float32)]\n",
      "\n",
      "running round 20\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.7379221, dtype=float32), DeviceArray(2.055602, dtype=float32), DeviceArray(1.9632256, dtype=float32)]\n",
      "0.4534856\n",
      "0.35473382\n",
      "0.2861763\n",
      "[1.0, DeviceArray(0.41177547, dtype=float32), DeviceArray(0.29306576, dtype=float32), DeviceArray(0.22727953, dtype=float32)]\n",
      "\n",
      "running round 21\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.9054527, dtype=float32), DeviceArray(1.9373792, dtype=float32), DeviceArray(2.2077851, dtype=float32)]\n",
      "0.41177547\n",
      "0.29306576\n",
      "0.22727953\n",
      "[1.0, DeviceArray(0.3420446, dtype=float32), DeviceArray(0.23106962, dtype=float32), DeviceArray(0.1566304, dtype=float32)]\n",
      "\n",
      "running round 22\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.4120395, dtype=float32), DeviceArray(2.1243665, dtype=float32), DeviceArray(2.0270162, dtype=float32)]\n",
      "0.3420446\n",
      "0.23106962\n",
      "0.1566304\n",
      "[1.0, DeviceArray(0.28415564, dtype=float32), DeviceArray(0.1630899, dtype=float32), DeviceArray(0.09176588, dtype=float32)]\n",
      "\n",
      "running round 23\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.8586974, dtype=float32), DeviceArray(2.0859962, dtype=float32), DeviceArray(2.1769989, dtype=float32)]\n",
      "0.28415564\n",
      "0.1630899\n",
      "0.09176588\n",
      "[1.0, DeviceArray(0.2358295, dtype=float32), DeviceArray(0.09216604, dtype=float32), DeviceArray(0.01774792, dtype=float32)]\n",
      "\n",
      "running round 24\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(3.6378555, dtype=float32), DeviceArray(2.7189429, dtype=float32), DeviceArray(2.5917015, dtype=float32)]\n",
      "0.2358295\n",
      "0.09216604\n",
      "0.017747924\n",
      "[1.0, DeviceArray(0.14124526, dtype=float32), DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 25\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.937289, dtype=float32), DeviceArray(2.500652, dtype=float32), DeviceArray(1.9587995, dtype=float32)]\n",
      "0.14124526\n",
      "0.0\n",
      "0.0\n",
      "[1.0, DeviceArray(0.06487575, dtype=float32), DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 26\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.7167637, dtype=float32), DeviceArray(3.8190877, dtype=float32), DeviceArray(3.3982975, dtype=float32)]\n",
      "0.06487575\n",
      "0.0\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 27\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(5.810923, dtype=float32), DeviceArray(4.116111, dtype=float32), DeviceArray(2.5016847, dtype=float32)]\n",
      "0.0\n",
      "0.0\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 28\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(4.9806657, dtype=float32), DeviceArray(6.230333, dtype=float32), DeviceArray(4.8126163, dtype=float32)]\n",
      "0.0\n",
      "0.0\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 29\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(4.1643043, dtype=float32), DeviceArray(5.976898, dtype=float32), DeviceArray(2.7765458, dtype=float32)]\n",
      "0.0\n",
      "0.0\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "\n",
      " --------------- TRAINING WITH Local (lr=3.0) AGGREGATOR FOR REQUESTING CLIENT = 0 ---------------\n",
      "\n",
      "running round 0\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "compiling: average_params\n",
      "num_params: 1\n",
      "\n",
      "running round 1\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "compiling: average_params\n",
      "num_params: 1\n",
      "\n",
      "running round 2\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 3\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 4\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 5\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 6\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 7\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 8\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 9\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 10\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 11\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 12\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 13\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 14\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 15\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 16\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 17\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 18\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 19\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 20\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 21\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 22\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 23\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 24\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 25\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 26\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 27\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 28\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 29\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "compiling: baseline_loss\n",
      "compiling: baseline_loss\n",
      "CLIENT 1 baseline test accuracy: 0.577\n",
      "\n",
      "\n",
      " --------------- TRAINING WITH Global (lr=0.07) AGGREGATOR FOR REQUESTING CLIENT = 1 ---------------\n",
      "compiling: loss\n",
      "compiling: accuracy\n",
      "Traced<ShapedArray(int32[78]):JaxprTrace(level=-1/1)>\n",
      "\n",
      "running round 0\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "\n",
      "running round 1\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "compiling: loss\n",
      "compiling: loss\n",
      "compiling: accuracy\n",
      "Traced<ShapedArray(int32[76]):JaxprTrace(level=-1/1)>\n",
      "compiling: run_one_step\n",
      "compiling: loss\n",
      "compiling: loss\n",
      "compiling: accuracy\n",
      "Traced<ShapedArray(int32[87]):JaxprTrace(level=-1/1)>\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "\n",
      "running round 2\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 3\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 4\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 5\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 6\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 7\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 8\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 9\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 10\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 11\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 12\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 13\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 14\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 15\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 16\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 17\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 18\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 19\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 20\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 21\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 22\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 23\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 24\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 25\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 26\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 27\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 28\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 29\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "\n",
      " --------------- TRAINING WITH Weight Erosion (lr=3.0) AGGREGATOR FOR REQUESTING CLIENT = 1 ---------------\n",
      "\n",
      "running round 0\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.5837776, dtype=float32), DeviceArray(0.48740247, dtype=float32), DeviceArray(0.7498417, dtype=float32)]\n",
      "1.0\n",
      "1.0\n",
      "1.0\n",
      "[1.0, DeviceArray(0.9941622, dtype=float32), DeviceArray(0.99512595, dtype=float32), DeviceArray(0.99250156, dtype=float32)]\n",
      "\n",
      "running round 1\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "compiling: run_one_step\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.9339705, dtype=float32), DeviceArray(1.0622908, dtype=float32), DeviceArray(1.8971634, dtype=float32)]\n",
      "0.9941622\n",
      "0.99512595\n",
      "0.99250156\n",
      "[1.0, DeviceArray(0.97095454, dtype=float32), DeviceArray(0.9823785, dtype=float32), DeviceArray(0.9697356, dtype=float32)]\n",
      "\n",
      "running round 2\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.8497593, dtype=float32), DeviceArray(0.4879835, dtype=float32), DeviceArray(1.0217905, dtype=float32)]\n",
      "0.97095454\n",
      "0.9823785\n",
      "0.9697356\n",
      "[1.0, DeviceArray(0.96075743, dtype=float32), DeviceArray(0.9765227, dtype=float32), DeviceArray(0.9574741, dtype=float32)]\n",
      "\n",
      "running round 3\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.8076671, dtype=float32), DeviceArray(1.4580649, dtype=float32), DeviceArray(2.259553, dtype=float32)]\n",
      "0.96075743\n",
      "0.9765227\n",
      "0.9574741\n",
      "[1.0, DeviceArray(0.9354501, dtype=float32), DeviceArray(0.95610976, dtype=float32), DeviceArray(0.9258404, dtype=float32)]\n",
      "\n",
      "running round 4\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.7305303, dtype=float32), DeviceArray(0.5186911, dtype=float32), DeviceArray(1.0292785, dtype=float32)]\n",
      "0.9354501\n",
      "0.95610976\n",
      "0.9258404\n",
      "[1.0, DeviceArray(0.92522264, dtype=float32), DeviceArray(0.94884807, dtype=float32), DeviceArray(0.9114305, dtype=float32)]\n",
      "\n",
      "running round 5\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.9177657, dtype=float32), DeviceArray(1.4792018, dtype=float32), DeviceArray(2.2793806, dtype=float32)]\n",
      "0.92522264\n",
      "0.94884807\n",
      "0.9114305\n",
      "[1.0, DeviceArray(0.8945384, dtype=float32), DeviceArray(0.92518085, dtype=float32), DeviceArray(0.87496036, dtype=float32)]\n",
      "\n",
      "running round 6\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.83077407, dtype=float32), DeviceArray(0.5389254, dtype=float32), DeviceArray(1.0723029, dtype=float32)]\n",
      "0.8945384\n",
      "0.92518085\n",
      "0.87496036\n",
      "[1.0, DeviceArray(0.88124603, dtype=float32), DeviceArray(0.916558, dtype=float32), DeviceArray(0.8578035, dtype=float32)]\n",
      "\n",
      "running round 7\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.09596, dtype=float32), DeviceArray(1.5123863, dtype=float32), DeviceArray(2.2457302, dtype=float32)]\n",
      "0.88124603\n",
      "0.916558\n",
      "0.8578035\n",
      "[1.0, DeviceArray(0.84351873, dtype=float32), DeviceArray(0.8893351, dtype=float32), DeviceArray(0.81738037, dtype=float32)]\n",
      "\n",
      "running round 8\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.92356807, dtype=float32), DeviceArray(0.56273484, dtype=float32), DeviceArray(1.1227579, dtype=float32)]\n",
      "0.84351873\n",
      "0.8893351\n",
      "0.81738037\n",
      "[1.0, DeviceArray(0.8268945, dtype=float32), DeviceArray(0.8792059, dtype=float32), DeviceArray(0.7971707, dtype=float32)]\n",
      "\n",
      "running round 9\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.2530386, dtype=float32), DeviceArray(1.5283498, dtype=float32), DeviceArray(2.177408, dtype=float32)]\n",
      "0.8268945\n",
      "0.8792059\n",
      "0.7971707\n",
      "[1.0, DeviceArray(0.78183377, dtype=float32), DeviceArray(0.8486389, dtype=float32), DeviceArray(0.75362253, dtype=float32)]\n",
      "\n",
      "running round 10\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.9972527, dtype=float32), DeviceArray(0.59048975, dtype=float32), DeviceArray(1.1771364, dtype=float32)]\n",
      "0.78183377\n",
      "0.8486389\n",
      "0.75362253\n",
      "[1.0, DeviceArray(0.76188874, dtype=float32), DeviceArray(0.8368291, dtype=float32), DeviceArray(0.7300798, dtype=float32)]\n",
      "\n",
      "running round 11\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.3725383, dtype=float32), DeviceArray(1.5303341, dtype=float32), DeviceArray(2.0925627, dtype=float32)]\n",
      "0.76188874\n",
      "0.8368291\n",
      "0.7300798\n",
      "[1.0, DeviceArray(0.7096929, dtype=float32), DeviceArray(0.8031618, dtype=float32), DeviceArray(0.68404347, dtype=float32)]\n",
      "\n",
      "running round 12\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.0568821, dtype=float32), DeviceArray(0.62065756, dtype=float32), DeviceArray(1.2327188, dtype=float32)]\n",
      "0.7096929\n",
      "0.8031618\n",
      "0.68404347\n",
      "[1.0, DeviceArray(0.6864415, dtype=float32), DeviceArray(0.7895073, dtype=float32), DeviceArray(0.65692365, dtype=float32)]\n",
      "\n",
      "running round 13\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.4605403, dtype=float32), DeviceArray(1.5269551, dtype=float32), DeviceArray(2.0063653, dtype=float32)]\n",
      "0.6864415\n",
      "0.7895073\n",
      "0.65692365\n",
      "[1.0, DeviceArray(0.62738854, dtype=float32), DeviceArray(0.7528604, dtype=float32), DeviceArray(0.6087709, dtype=float32)]\n",
      "\n",
      "running round 14\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.1057657, dtype=float32), DeviceArray(0.6527101, dtype=float32), DeviceArray(1.2876159, dtype=float32)]\n",
      "0.62738854\n",
      "0.7528604\n",
      "0.6087709\n",
      "[1.0, DeviceArray(0.60085016, dtype=float32), DeviceArray(0.7371954, dtype=float32), DeviceArray(0.5778681, dtype=float32)]\n",
      "\n",
      "running round 15\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.528896, dtype=float32), DeviceArray(1.5248892, dtype=float32), DeviceArray(1.928047, dtype=float32)]\n",
      "0.60085016\n",
      "0.7371954\n",
      "0.5778681\n",
      "[1.0, DeviceArray(0.53509885, dtype=float32), DeviceArray(0.6975483, dtype=float32), DeviceArray(0.52773887, dtype=float32)]\n",
      "\n",
      "running round 16\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.1458111, dtype=float32), DeviceArray(0.68756986, dtype=float32), DeviceArray(1.3405277, dtype=float32)]\n",
      "0.53509885\n",
      "0.6975483\n",
      "0.52773887\n",
      "[1.0, DeviceArray(0.5053078, dtype=float32), DeviceArray(0.67967147, dtype=float32), DeviceArray(0.49288514, dtype=float32)]\n",
      "\n",
      "running round 17\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.5895724, dtype=float32), DeviceArray(1.5285426, dtype=float32), DeviceArray(1.8619524, dtype=float32)]\n",
      "0.5053078\n",
      "0.67967147\n",
      "0.49288514\n",
      "[1.0, DeviceArray(0.43279976, dtype=float32), DeviceArray(0.6368723, dtype=float32), DeviceArray(0.44075048, dtype=float32)]\n",
      "\n",
      "running round 18\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.1786568, dtype=float32), DeviceArray(0.72839487, dtype=float32), DeviceArray(1.3904912, dtype=float32)]\n",
      "0.43279976\n",
      "0.6368723\n",
      "0.44075048\n",
      "[1.0, DeviceArray(0.39979738, dtype=float32), DeviceArray(0.61647725, dtype=float32), DeviceArray(0.40181673, dtype=float32)]\n",
      "\n",
      "running round 19\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.6547892, dtype=float32), DeviceArray(1.5415559, dtype=float32), DeviceArray(1.8098115, dtype=float32)]\n",
      "0.39979738\n",
      "0.61647725\n",
      "0.40181673\n",
      "[1.0, DeviceArray(0.3201537, dtype=float32), DeviceArray(0.5702306, dtype=float32), DeviceArray(0.34752238, dtype=float32)]\n",
      "\n",
      "running round 20\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.2071408, dtype=float32), DeviceArray(0.7826876, dtype=float32), DeviceArray(1.436404, dtype=float32)]\n",
      "0.3201537\n",
      "0.5702306\n",
      "0.34752238\n",
      "[1.0, DeviceArray(0.28393948, dtype=float32), DeviceArray(0.54674995, dtype=float32), DeviceArray(0.30443025, dtype=float32)]\n",
      "\n",
      "running round 21\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.7412467, dtype=float32), DeviceArray(1.569396, dtype=float32), DeviceArray(1.7729527, dtype=float32)]\n",
      "0.28393948\n",
      "0.54674995\n",
      "0.30443025\n",
      "[1.0, DeviceArray(0.1962196, dtype=float32), DeviceArray(0.49652928, dtype=float32), DeviceArray(0.24769576, dtype=float32)]\n",
      "\n",
      "running round 22\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.2394491, dtype=float32), DeviceArray(0.86744285, dtype=float32), DeviceArray(1.4757079, dtype=float32)]\n",
      "0.1962196\n",
      "0.49652928\n",
      "0.24769576\n",
      "[1.0, DeviceArray(0.15655722, dtype=float32), DeviceArray(0.4687711, dtype=float32), DeviceArray(0.2004731, dtype=float32)]\n",
      "\n",
      "running round 23\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.8831642, dtype=float32), DeviceArray(1.625792, dtype=float32), DeviceArray(1.7555171, dtype=float32)]\n",
      "0.15655722\n",
      "0.4687711\n",
      "0.2004731\n",
      "[1.0, DeviceArray(0.05852964, dtype=float32), DeviceArray(0.41349417, dtype=float32), DeviceArray(0.14078552, dtype=float32)]\n",
      "\n",
      "running round 24\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.3059131, dtype=float32), DeviceArray(1.0221754, dtype=float32), DeviceArray(1.4997717, dtype=float32)]\n",
      "0.058529638\n",
      "0.41349417\n",
      "0.14078552\n",
      "[1.0, DeviceArray(0.01412859, dtype=float32), DeviceArray(0.3787402, dtype=float32), DeviceArray(0.08979328, dtype=float32)]\n",
      "\n",
      "running round 25\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(3.1852028, dtype=float32), DeviceArray(1.7600769, dtype=float32), DeviceArray(1.7751735, dtype=float32)]\n",
      "0.014128592\n",
      "0.3787402\n",
      "0.08979328\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.3153774, dtype=float32), DeviceArray(0.02588703, dtype=float32)]\n",
      "\n",
      "running round 26\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.4717711, dtype=float32), DeviceArray(1.2998458, dtype=float32), DeviceArray(1.5492147, dtype=float32)]\n",
      "0.0\n",
      "0.3153774\n",
      "0.025887035\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.26858297, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 27\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(3.477566, dtype=float32), DeviceArray(1.9387612, dtype=float32), DeviceArray(1.7671584, dtype=float32)]\n",
      "0.0\n",
      "0.26858297\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.19491005, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 28\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.6815165, dtype=float32), DeviceArray(1.5885439, dtype=float32), DeviceArray(1.5574551, dtype=float32)]\n",
      "0.0\n",
      "0.19491005\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.13454539, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 29\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(3.691112, dtype=float32), DeviceArray(2.1906905, dtype=float32), DeviceArray(1.7116736, dtype=float32)]\n",
      "0.0\n",
      "0.13454539\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.04691777, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "\n",
      " --------------- TRAINING WITH Local (lr=3.0) AGGREGATOR FOR REQUESTING CLIENT = 1 ---------------\n",
      "\n",
      "running round 0\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "\n",
      "running round 1\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "\n",
      "running round 2\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 3\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 4\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 5\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 6\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 7\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 8\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 9\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 10\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 11\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 12\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 13\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 14\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 15\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 16\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 17\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 18\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 19\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 20\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 21\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 22\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 23\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 24\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 25\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 26\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 27\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 28\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 29\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "compiling: baseline_loss\n",
      "CLIENT 2 baseline test accuracy: 0.599\n",
      "\n",
      "\n",
      " --------------- TRAINING WITH Global (lr=0.07) AGGREGATOR FOR REQUESTING CLIENT = 2 ---------------\n",
      "compiling: loss\n",
      "compiling: accuracy\n",
      "Traced<ShapedArray(int32[1]):JaxprTrace(level=-1/1)>\n",
      "\n",
      "running round 0\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "compiling: loss\n",
      "compiling: loss\n",
      "compiling: accuracy\n",
      "Traced<ShapedArray(int32[160]):JaxprTrace(level=-1/1)>\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "\n",
      "running round 1\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "\n",
      "running round 2\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "\n",
      "running round 3\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 4\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 5\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 6\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 7\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 8\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 9\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 10\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 11\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 12\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 13\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 14\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 15\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 16\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 17\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 18\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 19\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 20\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 21\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 22\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 23\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 24\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 25\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 26\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 27\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 28\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 29\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "\n",
      " --------------- TRAINING WITH Weight Erosion (lr=3.0) AGGREGATOR FOR REQUESTING CLIENT = 2 ---------------\n",
      "\n",
      "running round 0\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.6497671, dtype=float32), DeviceArray(0.4355901, dtype=float32), DeviceArray(0.9380543, dtype=float32)]\n",
      "1.0\n",
      "1.0\n",
      "1.0\n",
      "[1.0, DeviceArray(0.9935023, dtype=float32), DeviceArray(0.9956441, dtype=float32), DeviceArray(0.9906195, dtype=float32)]\n",
      "\n",
      "running round 1\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.1308246, dtype=float32), DeviceArray(0.713458, dtype=float32), DeviceArray(1.5842929, dtype=float32)]\n",
      "0.9935023\n",
      "0.9956441\n",
      "0.9906195\n",
      "[1.0, DeviceArray(0.9799324, dtype=float32), DeviceArray(0.98850954, dtype=float32), DeviceArray(0.971608, dtype=float32)]\n",
      "\n",
      "running round 2\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.7128912, dtype=float32), DeviceArray(0.5903444, dtype=float32), DeviceArray(1.0245775, dtype=float32)]\n",
      "0.9799324\n",
      "0.98850954\n",
      "0.971608\n",
      "[1.0, DeviceArray(0.97137773, dtype=float32), DeviceArray(0.9814254, dtype=float32), DeviceArray(0.95931304, dtype=float32)]\n",
      "\n",
      "running round 3\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.2793638, dtype=float32), DeviceArray(0.99123555, dtype=float32), DeviceArray(2.2212684, dtype=float32)]\n",
      "0.97137773\n",
      "0.9814254\n",
      "0.95931304\n",
      "[1.0, DeviceArray(0.95346665, dtype=float32), DeviceArray(0.9695306, dtype=float32), DeviceArray(0.92821527, dtype=float32)]\n",
      "\n",
      "running round 4\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.833473, dtype=float32), DeviceArray(0.6779994, dtype=float32), DeviceArray(1.097973, dtype=float32)]\n",
      "0.95346665\n",
      "0.9695306\n",
      "0.92821527\n",
      "[1.0, DeviceArray(0.94179803, dtype=float32), DeviceArray(0.9613946, dtype=float32), DeviceArray(0.91284364, dtype=float32)]\n",
      "\n",
      "running round 5\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.4099053, dtype=float32), DeviceArray(0.9828824, dtype=float32), DeviceArray(2.373962, dtype=float32)]\n",
      "0.94179803\n",
      "0.9613946\n",
      "0.91284364\n",
      "[1.0, DeviceArray(0.9192395, dtype=float32), DeviceArray(0.9476343, dtype=float32), DeviceArray(0.8748602, dtype=float32)]\n",
      "\n",
      "running round 6\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.79783475, dtype=float32), DeviceArray(0.81510955, dtype=float32), DeviceArray(0.97238773, dtype=float32)]\n",
      "0.9192395\n",
      "0.9476343\n",
      "0.8748602\n",
      "[1.0, DeviceArray(0.9064742, dtype=float32), DeviceArray(0.93622273, dtype=float32), DeviceArray(0.85930204, dtype=float32)]\n",
      "\n",
      "running round 7\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.5359236, dtype=float32), DeviceArray(1.031743, dtype=float32), DeviceArray(2.4160688, dtype=float32)]\n",
      "0.9064742\n",
      "0.93622273\n",
      "0.85930204\n",
      "[1.0, DeviceArray(0.8788276, dtype=float32), DeviceArray(0.9217783, dtype=float32), DeviceArray(0.8158128, dtype=float32)]\n",
      "\n",
      "running round 8\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.1888607, dtype=float32), DeviceArray(0.7132973, dtype=float32), DeviceArray(1.1652517, dtype=float32)]\n",
      "0.8788276\n",
      "0.9217783\n",
      "0.8158128\n",
      "[1.0, DeviceArray(0.8574281, dtype=float32), DeviceArray(0.9103656, dtype=float32), DeviceArray(0.7948383, dtype=float32)]\n",
      "\n",
      "running round 9\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.9038587, dtype=float32), DeviceArray(1.2936656, dtype=float32), DeviceArray(2.5944166, dtype=float32)]\n",
      "0.8574281\n",
      "0.9103656\n",
      "0.7948383\n",
      "[1.0, DeviceArray(0.8193509, dtype=float32), DeviceArray(0.8896669, dtype=float32), DeviceArray(0.74294996, dtype=float32)]\n",
      "\n",
      "running round 10\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.4366045, dtype=float32), DeviceArray(0.81407833, dtype=float32), DeviceArray(1.275458, dtype=float32)]\n",
      "0.8193509\n",
      "0.8896669\n",
      "0.74294996\n",
      "[1.0, DeviceArray(0.79061884, dtype=float32), DeviceArray(0.8766417, dtype=float32), DeviceArray(0.7174408, dtype=float32)]\n",
      "\n",
      "running round 11\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.925691, dtype=float32), DeviceArray(1.2035223, dtype=float32), DeviceArray(2.7344155, dtype=float32)]\n",
      "0.79061884\n",
      "0.8766417\n",
      "0.7174408\n",
      "[1.0, DeviceArray(0.74825364, dtype=float32), DeviceArray(0.85497826, dtype=float32), DeviceArray(0.65728366, dtype=float32)]\n",
      "\n",
      "running round 12\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.1083002, dtype=float32), DeviceArray(1.0228316, dtype=float32), DeviceArray(1.1329454, dtype=float32)]\n",
      "0.74825364\n",
      "0.85497826\n",
      "0.65728366\n",
      "[1.0, DeviceArray(0.72387105, dtype=float32), DeviceArray(0.8365673, dtype=float32), DeviceArray(0.63235885, dtype=float32)]\n",
      "\n",
      "running round 13\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.9344074, dtype=float32), DeviceArray(1.1870328, dtype=float32), DeviceArray(2.6596758, dtype=float32)]\n",
      "0.72387105\n",
      "0.8365673\n",
      "0.63235885\n",
      "[1.0, DeviceArray(0.6774453, dtype=float32), DeviceArray(0.8152007, dtype=float32), DeviceArray(0.5685266, dtype=float32)]\n",
      "\n",
      "running round 14\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.6719861, dtype=float32), DeviceArray(1.0707297, dtype=float32), DeviceArray(1.4030169, dtype=float32)]\n",
      "0.6774453\n",
      "0.8152007\n",
      "0.5685266\n",
      "[1.0, DeviceArray(0.6373176, dtype=float32), DeviceArray(0.7937861, dtype=float32), DeviceArray(0.53485423, dtype=float32)]\n",
      "\n",
      "running round 15\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.5576494, dtype=float32), DeviceArray(1.6147975, dtype=float32), DeviceArray(2.9843445, dtype=float32)]\n",
      "0.6373176\n",
      "0.7937861\n",
      "0.53485423\n",
      "[1.0, DeviceArray(0.5708187, dtype=float32), DeviceArray(0.76149017, dtype=float32), DeviceArray(0.45726126, dtype=float32)]\n",
      "\n",
      "running round 16\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.0176218, dtype=float32), DeviceArray(1.0624945, dtype=float32), DeviceArray(1.5865965, dtype=float32)]\n",
      "0.5708187\n",
      "0.76149017\n",
      "0.45726126\n",
      "[1.0, DeviceArray(0.51836056, dtype=float32), DeviceArray(0.7402403, dtype=float32), DeviceArray(0.41600975, dtype=float32)]\n",
      "\n",
      "running round 17\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.4751515, dtype=float32), DeviceArray(1.5123196, dtype=float32), DeviceArray(3.2041647, dtype=float32)]\n",
      "0.51836056\n",
      "0.7402403\n",
      "0.41600975\n",
      "[1.0, DeviceArray(0.44905633, dtype=float32), DeviceArray(0.70696926, dtype=float32), DeviceArray(0.32629314, dtype=float32)]\n",
      "\n",
      "running round 18\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.4556553, dtype=float32), DeviceArray(1.2245084, dtype=float32), DeviceArray(1.3763047, dtype=float32)]\n",
      "0.44905633\n",
      "0.70696926\n",
      "0.32629314\n",
      "[1.0, DeviceArray(0.408298, dtype=float32), DeviceArray(0.68003005, dtype=float32), DeviceArray(0.28775662, dtype=float32)]\n",
      "\n",
      "running round 19\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.2182124, dtype=float32), DeviceArray(1.3050047, dtype=float32), DeviceArray(2.82722, dtype=float32)]\n",
      "0.408298\n",
      "0.68003005\n",
      "0.28775662\n",
      "[1.0, DeviceArray(0.3417516, dtype=float32), DeviceArray(0.6513199, dtype=float32), DeviceArray(0.20294002, dtype=float32)]\n",
      "\n",
      "running round 20\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.5024288, dtype=float32), DeviceArray(1.6809394, dtype=float32), DeviceArray(1.8263499, dtype=float32)]\n",
      "0.3417516\n",
      "0.6513199\n",
      "0.20294002\n",
      "[1.0, DeviceArray(0.26667875, dtype=float32), DeviceArray(0.61097735, dtype=float32), DeviceArray(0.14814952, dtype=float32)]\n",
      "\n",
      "running round 21\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(3.197399, dtype=float32), DeviceArray(1.861333, dtype=float32), DeviceArray(3.589682, dtype=float32)]\n",
      "0.26667875\n",
      "0.61097735\n",
      "0.14814952\n",
      "[1.0, DeviceArray(0.16436198, dtype=float32), DeviceArray(0.56630534, dtype=float32), DeviceArray(0.03327969, dtype=float32)]\n",
      "\n",
      "running round 22\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.7455692, dtype=float32), DeviceArray(1.390841, dtype=float32), DeviceArray(1.8161343, dtype=float32)]\n",
      "0.16436198\n",
      "0.56630534\n",
      "0.033279695\n",
      "[1.0, DeviceArray(0.07650377, dtype=float32), DeviceArray(0.5329251, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 23\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(3.5244381, dtype=float32), DeviceArray(2.380682, dtype=float32), DeviceArray(4.589748, dtype=float32)]\n",
      "0.07650377\n",
      "0.5329251\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.4710274, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 24\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.9270422, dtype=float32), DeviceArray(1.7393885, dtype=float32), DeviceArray(2.1236668, dtype=float32)]\n",
      "0.0\n",
      "0.4710274\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.4258033, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 25\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.8744433, dtype=float32), DeviceArray(1.4335526, dtype=float32), DeviceArray(3.683847, dtype=float32)]\n",
      "0.0\n",
      "0.4258033\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.38853094, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 26\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(4.4858336, dtype=float32), DeviceArray(2.9946074, dtype=float32), DeviceArray(2.6889365, dtype=float32)]\n",
      "0.0\n",
      "0.38853094\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.30468193, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 27\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(4.400856, dtype=float32), DeviceArray(2.0280166, dtype=float32), DeviceArray(5.393654, dtype=float32)]\n",
      "0.0\n",
      "0.30468193\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.24789746, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 28\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(4.3838634, dtype=float32), DeviceArray(1.7453667, dtype=float32), DeviceArray(2.5492551, dtype=float32)]\n",
      "0.0\n",
      "0.24789746\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.1990272, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 29\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(5.022679, dtype=float32), DeviceArray(3.57347, dtype=float32), DeviceArray(6.303105, dtype=float32)]\n",
      "0.0\n",
      "0.1990272\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.09182309, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "\n",
      " --------------- TRAINING WITH Local (lr=3.0) AGGREGATOR FOR REQUESTING CLIENT = 2 ---------------\n",
      "\n",
      "running round 0\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "\n",
      "running round 1\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 2\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 3\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 4\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 5\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 6\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 7\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 8\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 9\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 10\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 11\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 12\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 13\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 14\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 15\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 16\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 17\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 18\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 19\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 20\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 21\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 22\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 23\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 24\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 25\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 26\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 27\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 28\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 29\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "compiling: baseline_loss\n",
      "CLIENT 3 baseline test accuracy: 0.689\n",
      "\n",
      "\n",
      " --------------- TRAINING WITH Global (lr=0.07) AGGREGATOR FOR REQUESTING CLIENT = 3 ---------------\n",
      "compiling: loss\n",
      "compiling: accuracy\n",
      "Traced<ShapedArray(int32[132]):JaxprTrace(level=-1/1)>\n",
      "\n",
      "running round 0\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "compiling: loss\n",
      "compiling: loss\n",
      "compiling: accuracy\n",
      "Traced<ShapedArray(int32[131]):JaxprTrace(level=-1/1)>\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "\n",
      "running round 1\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "\n",
      "running round 2\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "\n",
      "running round 3\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 4\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 5\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 6\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 7\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 8\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 9\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 10\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 11\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 12\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 13\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 14\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 15\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 16\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 17\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 18\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 19\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 20\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 21\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 22\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 23\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 24\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 25\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 26\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 27\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 28\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 29\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "\n",
      " --------------- TRAINING WITH Weight Erosion (lr=3.0) AGGREGATOR FOR REQUESTING CLIENT = 3 ---------------\n",
      "\n",
      "running round 0\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(0.90666574, dtype=float32), DeviceArray(0.8139351, dtype=float32), DeviceArray(0.8404179, dtype=float32)]\n",
      "1.0\n",
      "1.0\n",
      "1.0\n",
      "[1.0, DeviceArray(0.99093336, dtype=float32), DeviceArray(0.9918606, dtype=float32), DeviceArray(0.9915958, dtype=float32)]\n",
      "\n",
      "running round 1\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.872789, dtype=float32), DeviceArray(1.5521449, dtype=float32), DeviceArray(1.178583, dtype=float32)]\n",
      "0.99093336\n",
      "0.9918606\n",
      "0.9915958\n",
      "[1.0, DeviceArray(0.9684599, dtype=float32), DeviceArray(0.97633916, dtype=float32), DeviceArray(0.9774528, dtype=float32)]\n",
      "\n",
      "running round 2\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.3302929, dtype=float32), DeviceArray(1.0090953, dtype=float32), DeviceArray(1.1309141, dtype=float32)]\n",
      "0.9684599\n",
      "0.97633916\n",
      "0.9774528\n",
      "[1.0, DeviceArray(0.9524964, dtype=float32), DeviceArray(0.96423, dtype=float32), DeviceArray(0.96388185, dtype=float32)]\n",
      "\n",
      "running round 3\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.8260279, dtype=float32), DeviceArray(1.8453441, dtype=float32), DeviceArray(1.2244352, dtype=float32)]\n",
      "0.9524964\n",
      "0.96423\n",
      "0.96388185\n",
      "[1.0, DeviceArray(0.92693204, dtype=float32), DeviceArray(0.94208586, dtype=float32), DeviceArray(0.94673973, dtype=float32)]\n",
      "\n",
      "running round 4\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.1228032, dtype=float32), DeviceArray(1.5495714, dtype=float32), DeviceArray(1.2593873, dtype=float32)]\n",
      "0.92693204\n",
      "0.94208586\n",
      "0.94673973\n",
      "[1.0, DeviceArray(0.9112128, dtype=float32), DeviceArray(0.923491, dtype=float32), DeviceArray(0.9291083, dtype=float32)]\n",
      "\n",
      "running round 5\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.4803188, dtype=float32), DeviceArray(1.0642117, dtype=float32), DeviceArray(1.1913991, dtype=float32)]\n",
      "0.9112128\n",
      "0.923491\n",
      "0.9291083\n",
      "[1.0, DeviceArray(0.8875277, dtype=float32), DeviceArray(0.90859205, dtype=float32), DeviceArray(0.9100459, dtype=float32)]\n",
      "\n",
      "running round 6\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.3033708, dtype=float32), DeviceArray(2.109496, dtype=float32), DeviceArray(1.5770938, dtype=float32)]\n",
      "0.8875277\n",
      "0.90859205\n",
      "0.9100459\n",
      "[1.0, DeviceArray(0.86667377, dtype=float32), DeviceArray(0.8790591, dtype=float32), DeviceArray(0.8848124, dtype=float32)]\n",
      "\n",
      "running round 7\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.4933519, dtype=float32), DeviceArray(1.6314738, dtype=float32), DeviceArray(1.1892068, dtype=float32)]\n",
      "0.86667377\n",
      "0.8790591\n",
      "0.8848124\n",
      "[1.0, DeviceArray(0.83979344, dtype=float32), DeviceArray(0.85621846, dtype=float32), DeviceArray(0.8634067, dtype=float32)]\n",
      "\n",
      "running round 8\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.0704728, dtype=float32), DeviceArray(1.057715, dtype=float32), DeviceArray(1.389164, dtype=float32)]\n",
      "0.83979344\n",
      "0.85621846\n",
      "0.8634067\n",
      "[1.0, DeviceArray(0.82052493, dtype=float32), DeviceArray(0.839295, dtype=float32), DeviceArray(0.8384018, dtype=float32)]\n",
      "\n",
      "running round 9\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.1302154, dtype=float32), DeviceArray(2.339563, dtype=float32), DeviceArray(1.4048395, dtype=float32)]\n",
      "0.82052493\n",
      "0.839295\n",
      "0.8384018\n",
      "[1.0, DeviceArray(0.7779206, dtype=float32), DeviceArray(0.801862, dtype=float32), DeviceArray(0.810305, dtype=float32)]\n",
      "\n",
      "running round 10\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.0301399, dtype=float32), DeviceArray(1.6597195, dtype=float32), DeviceArray(1.3534569, dtype=float32)]\n",
      "0.7779206\n",
      "0.801862\n",
      "0.810305\n",
      "[1.0, DeviceArray(0.7573178, dtype=float32), DeviceArray(0.77530646, dtype=float32), DeviceArray(0.78323585, dtype=float32)]\n",
      "\n",
      "running round 11\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.795881, dtype=float32), DeviceArray(1.0713899, dtype=float32), DeviceArray(1.2767137, dtype=float32)]\n",
      "0.7573178\n",
      "0.77530646\n",
      "0.78323585\n",
      "[1.0, DeviceArray(0.71780837, dtype=float32), DeviceArray(0.75602144, dtype=float32), DeviceArray(0.7551482, dtype=float32)]\n",
      "\n",
      "running round 12\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.3015794, dtype=float32), DeviceArray(2.4518301, dtype=float32), DeviceArray(1.7576596, dtype=float32)]\n",
      "0.71780837\n",
      "0.75602144\n",
      "0.7551482\n",
      "[1.0, DeviceArray(0.68917364, dtype=float32), DeviceArray(0.7118885, dtype=float32), DeviceArray(0.71647966, dtype=float32)]\n",
      "\n",
      "running round 13\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.8099558, dtype=float32), DeviceArray(1.7664281, dtype=float32), DeviceArray(1.2613647, dtype=float32)]\n",
      "0.68917364\n",
      "0.7118885\n",
      "0.71647966\n",
      "[1.0, DeviceArray(0.64573467, dtype=float32), DeviceArray(0.6800928, dtype=float32), DeviceArray(0.6862069, dtype=float32)]\n",
      "\n",
      "running round 14\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.0855603, dtype=float32), DeviceArray(1.0433607, dtype=float32), DeviceArray(1.5245768, dtype=float32)]\n",
      "0.64573467\n",
      "0.6800928\n",
      "0.6862069\n",
      "[1.0, DeviceArray(0.61968124, dtype=float32), DeviceArray(0.6592256, dtype=float32), DeviceArray(0.649617, dtype=float32)]\n",
      "\n",
      "running round 15\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.6108396, dtype=float32), DeviceArray(2.7285845, dtype=float32), DeviceArray(1.6460235, dtype=float32)]\n",
      "0.61968124\n",
      "0.6592256\n",
      "0.649617\n",
      "[1.0, DeviceArray(0.5517994, dtype=float32), DeviceArray(0.6046539, dtype=float32), DeviceArray(0.6068204, dtype=float32)]\n",
      "\n",
      "running round 16\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.0773827, dtype=float32), DeviceArray(1.7868018, dtype=float32), DeviceArray(1.486414, dtype=float32)]\n",
      "0.5517994\n",
      "0.6046539\n",
      "0.6068204\n",
      "[1.0, DeviceArray(0.5237875, dtype=float32), DeviceArray(0.5689179, dtype=float32), DeviceArray(0.56817365, dtype=float32)]\n",
      "\n",
      "running round 17\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.225308, dtype=float32), DeviceArray(1.1067625, dtype=float32), DeviceArray(1.4764594, dtype=float32)]\n",
      "0.5237875\n",
      "0.5689179\n",
      "0.56817365\n",
      "[1.0, DeviceArray(0.4614789, dtype=float32), DeviceArray(0.5445691, dtype=float32), DeviceArray(0.52683276, dtype=float32)]\n",
      "\n",
      "running round 18\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.4957381, dtype=float32), DeviceArray(2.8427916, dtype=float32), DeviceArray(2.042426, dtype=float32)]\n",
      "0.4614789\n",
      "0.5445691\n",
      "0.52683276\n",
      "[1.0, DeviceArray(0.41959822, dtype=float32), DeviceArray(0.48202765, dtype=float32), DeviceArray(0.46964484, dtype=float32)]\n",
      "\n",
      "running round 19\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(2.2916098, dtype=float32), DeviceArray(1.9877986, dtype=float32), DeviceArray(1.5051237, dtype=float32)]\n",
      "0.41959822\n",
      "0.48202765\n",
      "0.46964484\n",
      "[1.0, DeviceArray(0.35084993, dtype=float32), DeviceArray(0.43829608, dtype=float32), DeviceArray(0.42449114, dtype=float32)]\n",
      "\n",
      "running round 20\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.3337499, dtype=float32), DeviceArray(1.1403869, dtype=float32), DeviceArray(1.8251343, dtype=float32)]\n",
      "0.35084993\n",
      "0.43829608\n",
      "0.42449114\n",
      "[1.0, DeviceArray(0.31083745, dtype=float32), DeviceArray(0.4109268, dtype=float32), DeviceArray(0.36973712, dtype=float32)]\n",
      "\n",
      "running round 21\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(3.3953373, dtype=float32), DeviceArray(3.3025641, dtype=float32), DeviceArray(2.2259529, dtype=float32)]\n",
      "0.31083745\n",
      "0.4109268\n",
      "0.36973712\n",
      "[1.0, DeviceArray(0.20218664, dtype=float32), DeviceArray(0.33166525, dtype=float32), DeviceArray(0.29850662, dtype=float32)]\n",
      "\n",
      "running round 22\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(1.5983943, dtype=float32), DeviceArray(2.235428, dtype=float32), DeviceArray(1.923911, dtype=float32)]\n",
      "0.20218664\n",
      "0.33166525\n",
      "0.29850662\n",
      "[1.0, DeviceArray(0.15103802, dtype=float32), DeviceArray(0.27801496, dtype=float32), DeviceArray(0.23694147, dtype=float32)]\n",
      "\n",
      "running round 23\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(3.470534, dtype=float32), DeviceArray(1.4089861, dtype=float32), DeviceArray(2.2652142, dtype=float32)]\n",
      "0.15103802\n",
      "0.27801496\n",
      "0.23694147\n",
      "[1.0, DeviceArray(0.03303986, dtype=float32), DeviceArray(0.24138132, dtype=float32), DeviceArray(0.15992418, dtype=float32)]\n",
      "\n",
      "running round 24\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(3.321846, dtype=float32), DeviceArray(4.298498, dtype=float32), DeviceArray(3.124487, dtype=float32)]\n",
      "0.03303986\n",
      "0.24138132\n",
      "0.15992418\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.12962037, dtype=float32), DeviceArray(0.05369163, dtype=float32)]\n",
      "\n",
      "running round 25\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(5.7281623, dtype=float32), DeviceArray(3.7754385, dtype=float32), DeviceArray(3.6838944, dtype=float32)]\n",
      "0.0\n",
      "0.12962037\n",
      "0.053691626\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0.03145898, dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 26\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(6.05065, dtype=float32), DeviceArray(2.6757636, dtype=float32), DeviceArray(5.104936, dtype=float32)]\n",
      "0.0\n",
      "0.03145898\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 27\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(9.850857, dtype=float32), DeviceArray(7.4863725, dtype=float32), DeviceArray(7.3543363, dtype=float32)]\n",
      "0.0\n",
      "0.0\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 28\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(9.258825, dtype=float32), DeviceArray(7.2077265, dtype=float32), DeviceArray(7.629236, dtype=float32)]\n",
      "0.0\n",
      "0.0\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "running round 29\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "rel_distances_of_others: [DeviceArray(13.082991, dtype=float32), DeviceArray(4.1092205, dtype=float32), DeviceArray(10.147085, dtype=float32)]\n",
      "0.0\n",
      "0.0\n",
      "0.0\n",
      "[1.0, DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32), DeviceArray(0., dtype=float32)]\n",
      "\n",
      "\n",
      " --------------- TRAINING WITH Local (lr=3.0) AGGREGATOR FOR REQUESTING CLIENT = 3 ---------------\n",
      "\n",
      "running round 0\n",
      "computing updates from active clients.\n",
      "compiling: run_one_step\n",
      "aggregating client updates.\n",
      "\n",
      "running round 1\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 2\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 3\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 4\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 5\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 6\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 7\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 8\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 9\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 10\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 11\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 12\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 13\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 14\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 15\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 16\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 17\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 18\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 19\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 20\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 21\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 22\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 23\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 24\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 25\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 26\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 27\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 28\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n",
      "\n",
      "running round 29\n",
      "computing updates from active clients.\n",
      "aggregating client updates.\n"
     ]
    }
   ],
   "source": [
    "# testing federated learning code.\n",
    "\n",
    "hyperparams = ServerHyperParams(\n",
    "    num_rounds = 30,\n",
    "    max_batches_per_round = 1,\n",
    "    max_epochs_per_round = 1,\n",
    "    batch_size = 161,\n",
    "    seed = 278\n",
    ")\n",
    "\n",
    "sim_aggregator = init_similarity_aggregator()\n",
    "avg_aggregator = init_average_aggregator()\n",
    "\n",
    "constant_inputs = {\n",
    "        'hyperparams': hyperparams,\n",
    "        'server_opt': optix.sgd(1.0),\n",
    "        'loss': loss,\n",
    "        'init': init\n",
    "}\n",
    "\n",
    "set_loader = 'TITANIC_SURVIVED' # 'ANNIE_EVD' / 'TITANIC_SURVIVED'\n",
    "splitter   = 'AGE_STRICT' # 'AGE_STRICT' / 'AGE_SOME' / 'SINGLE_CLIENT' / 'PREDICT_AGE_SINGLE_CLIENT'\n",
    "\n",
    "\n",
    "random.seed(hyperparams.seed)\n",
    "client_sets = utils.load(utils.SetLoaders[set_loader], utils.Splitters[splitter], hyperparams.seed)\n",
    "diagnostics, model_params = {}, {}\n",
    "baseline_metrics = []\n",
    "for requesting_client in range(len(client_sets)):\n",
    "    dss, ds_test = make_dss(client_sets, hyperparams, requesting_client)\n",
    "    jax_ds_test  = make_jax_dataset(ds_test)\n",
    "    \n",
    "    training_types = {\n",
    "        'Global (lr={:.2f})'.format(lr): {'client_data': dss,\n",
    "                 'aggregator' : avg_aggregator,\n",
    "                 'client_opt' : optix.sgd(lr)\n",
    "                } for lr in [0.07] # np.logspace(-1.5, -0.2, num=5)\n",
    "    }\n",
    "    training_types.update({\n",
    "        'Weight Erosion (lr={:.1f})'.format(lr): {'client_data': dss,\n",
    "                 'aggregator' : sim_aggregator,\n",
    "                 'client_opt' : optix.sgd(lr)\n",
    "                } for lr in [3.]\n",
    "    })\n",
    "    training_types.update({\n",
    "        'Local (lr={:.1f})'.format(lr): {'client_data': dss[0:1],\n",
    "                 'aggregator' : avg_aggregator,\n",
    "                 'client_opt' : optix.sgd(lr)\n",
    "                } for lr in [3.0] # np.logspace(-1, 1, num=5)\n",
    "    })\n",
    "    \n",
    "    baseline_metrics.append(DiagnosticsMessage(\n",
    "            train_loss = None,\n",
    "            train_acc  = None,\n",
    "            test_loss  = eval_baseline_loss(ds_test),\n",
    "            test_acc   = eval_baseline_test_accuracy(ds_test),\n",
    "            weight     = None,\n",
    "            ))\n",
    "    print(\"CLIENT {} baseline test accuracy: {:.3f}\".format(requesting_client, baseline_metrics[-1].test_acc))\n",
    "\n",
    "    for key, varied_inputs in training_types.items():\n",
    "        print('\\n'*2, '-'*15, 'TRAINING WITH {} AGGREGATOR FOR REQUESTING CLIENT = {}'.format(key, requesting_client), '-'*15)\n",
    "        opt_params, \\\n",
    "        diag_state          = federated_learning(ds_test = jax_ds_test,\n",
    "                                                 **varied_inputs,\n",
    "                                                 **constant_inputs\n",
    "                                                );\n",
    "        if not key in diagnostics:\n",
    "            diagnostics[key]  = []\n",
    "            model_params[key] = []\n",
    "        diagnostics[ key].append(diag_state)\n",
    "        model_params[key].append(opt_params)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3\n"
     ]
    }
   ],
   "source": [
    "# requesting_client = 2\n",
    "print(requesting_client)\n",
    "# dss, ds_test = make_dss(client_sets, hyperparams, requesting_client)\n",
    "# for i in range(len(dss)):\n",
    "#     for batch in dss[i]:\n",
    "#         print(\"trainig batch\", i)\n",
    "# for batch in ds_test:\n",
    "#     print(\"test batch\")\n",
    "#     print(batch)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Display results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "markers = {k:v for k,v in zip(diagnostics.keys(), 'ox+svs'*8)}\n",
    "colors = {k:v for k,v in zip(diagnostics.keys(), 'gbkcm'*8)}\n",
    "\n",
    "def final_scatter(accuracies, **kwargs):\n",
    "    plt.scatter(np.arange(len(accuracies)), accuracies, s=60, **kwargs)\n",
    "\n",
    "final_scatter([dm.test_acc for dm in baseline_metrics], label='baseline test accuracy', marker='_', c='r')\n",
    "\n",
    "for key, val in diagnostics.items():\n",
    "    test_acc = [val[i].test_accuracy[-1] for i in range(len(val))]\n",
    "    x = np.arange(len(val))\n",
    "    final_scatter(test_acc, label=key, marker=markers[key], c=colors[key])\n",
    "    \n",
    "plt.legend()\n",
    "plt.xlabel('requesting client')\n",
    "plt.ylabel('accuracy on the test set')\n",
    "plt.title(\"Final test accuracy\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "def average_diag_states(dstates: Sequence[DiagnosticsState]) -> DiagnosticsState:\n",
    "    args = [np.mean([ d[i] for d in dstates], axis=0)\n",
    "            for i in range(len(dstates[0]))]\n",
    "    return DiagnosticsState(*args)\n",
    "    \n",
    "avg_diagnostics = {}\n",
    "for key, dlist in diagnostics.items():\n",
    "    avg_diagnostics[key] = average_diag_states(dlist)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Plot the evolution of test accuracy and client similarities during training\n",
    "def plot_test_accuracy_and_similarities(diags: Mapping[str, DiagnosticsState], baseline, title='', ax=None, plot_train_loss_instead=False):\n",
    "    make_fig = (ax is None)\n",
    "    if make_fig:\n",
    "        fig, ax = plt.subplots(figsize=(9,7))\n",
    "        ms = 6\n",
    "        fontsmall = 12\n",
    "        fontlarge = 15\n",
    "    else:\n",
    "        ms = 4\n",
    "        fontsmall = 14\n",
    "        fontlarge = 17\n",
    "    \n",
    "    \n",
    "    ax.set_xlabel(\"Communication round\", fontsize=fontsmall)\n",
    "#     ax2 = ax.twinx()\n",
    "\n",
    "    if plot_train_loss_instead:\n",
    "        h = ax.axhline(baseline.test_loss, color='r', linestyle='dashed')\n",
    "    else:\n",
    "        h = ax.axhline(baseline.test_acc, color='r', linestyle='dashed')\n",
    "        ax.axhline(1-baseline.test_acc, color='r', linestyle='dashed')\n",
    "    \n",
    "    handles_methods, labels_methods, handles_sim, labels_sim = [h], ['baseline accuracy'], [], []\n",
    "    for key, dstate in diags.items():\n",
    "        m = markers[key]\n",
    "        c = colors[key]\n",
    "        if plot_train_loss_instead:\n",
    "            h, = ax.plot(dstate.train_loss_global, '-'+m+c, ms=ms)\n",
    "        else:\n",
    "            h, = ax.plot(dstate.test_accuracy, '-'+m+c, ms=ms)\n",
    "        handles_methods.append(h)\n",
    "        labels_methods.append(key)\n",
    "        if key.lower().startswith('weight'):\n",
    "            handles_sim = ax.plot(dstate.similarities, ':')\n",
    "            num_clients = np.array(dstate.similarities).shape[1]\n",
    "#             ax2.plot(dstate.epoch_counts, ':')\n",
    "    if plot_train_loss_instead:\n",
    "        ax.set_ylabel(\"Agent weight / Loss\", fontsize=fontsmall)\n",
    "    else:\n",
    "        ax.set_ylabel(\"Agent weight / Test accuracy\", fontsize=fontsmall)\n",
    "        ax.set_ylim((-0.02,1.02))\n",
    "#     ax2.set_ylabel(\"Epoch count\")\n",
    "    handles = handles_methods + handles_sim\n",
    "    if handles_sim:\n",
    "        labels_sim = ['Weight agent {}'.format(i) for i in range(num_clients)]\n",
    "    labels = labels_methods + labels_sim\n",
    "    ax.set_title(title, fontsize=fontlarge)\n",
    "    if make_fig:\n",
    "        ax.legend(handles, labels, fontsize=fontsmall)\n",
    "        plt.show()\n",
    "        return\n",
    "    return handles, labels, fontsmall"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x504 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# averaged over the various requesting clients\n",
    "plot_test_accuracy_and_similarities(avg_diagnostics, baseline_metrics[0], \"Averaged over the users (agent ordering is lost)\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_diagnostics_for_requesting_client(dstates: Mapping[str, Sequence[DiagnosticsState]],\n",
    "                                          requesting_client: int\n",
    "                                         ) -> Mapping[str, DiagnosticsState]:\n",
    "    \"\"\"\n",
    "    INPUTS:\n",
    "        dstates: dict\n",
    "            key: data set selection strategy (e.g., 'global' or 'similar')\n",
    "            value: list of DiagnosticsState. Each DiagnosticsState\n",
    "                   corresponds to one requesting client.\n",
    "        requesting_client: int\n",
    "    \n",
    "    OUTPUTS:\n",
    "        res: dict\n",
    "            key: data set selection strategy (same keys as input `dstates`)\n",
    "            value: DiagnosticsState corresponding to client `requesting_client`.\n",
    "            \n",
    "    The DiagnosticsState.similarities and DiagnosticsState.epoch_counts are reordered,\n",
    "    to maintain the identity of each dataset. For example, the dataset\n",
    "    (, similarity, and epoch_count,) of client 1 are at index 1 irrespectively of\n",
    "    who the requesting client is. This is not the case in the raw training results.\n",
    "    \"\"\"\n",
    "    res = {}\n",
    "    for key, dlist in diagnostics.items():\n",
    "        dstate = dlist[requesting_client]\n",
    "        similarities = np.array(dstate.similarities)\n",
    "        num_clients = similarities.shape[1]\n",
    "        \n",
    "        indexes_reordered = list(range(1, requesting_client+1)) + [0] + list(range(requesting_client+1, num_clients))\n",
    "        if requesting_client >= num_clients:\n",
    "            indexes_reordered = list(range(num_clients))\n",
    "        \n",
    "        similarities_reordered = np.array(dstate.similarities).T[indexes_reordered].T\n",
    "        epoch_counts_reordered = np.array(dstate.epoch_counts).T[indexes_reordered].T\n",
    "#         print(num_clients, indexes_reordered)\n",
    "#         print(similarities_reordered)\n",
    "        res[key] = dstate._replace(similarities=similarities_reordered,\n",
    "                                   epoch_counts=epoch_counts_reordered\n",
    "                                  )\n",
    "    return res"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x432 with 6 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Individually per requesting client\n",
    "fig = plt.figure(figsize=(15,6))\n",
    "\n",
    "gs = fig.add_gridspec(2, len(client_sets), wspace=0.3, height_ratios=[1,4], hspace=0.3)\n",
    "axs = [fig.add_subplot(gs[1,i]) for i in range(len(baseline_metrics))]\n",
    "\n",
    "for requesting_client, ax in enumerate(axs):\n",
    "    handles, labels, fsize = plot_test_accuracy_and_similarities(\n",
    "        get_diagnostics_for_requesting_client(diagnostics, requesting_client),\n",
    "        baseline_metrics[requesting_client],\n",
    "        \"User: agent {}\".format(requesting_client),\n",
    "        ax=ax,\n",
    "        plot_train_loss_instead=False\n",
    "    )\n",
    "    \n",
    "n_accs = len(training_types) + 1\n",
    "include_weights = len(labels) > n_accs\n",
    "legend_kwargs = {'ncol':           4 - 2*include_weights,\n",
    "                 'fontsize':       fsize,\n",
    "                 'title_fontsize': fsize+1,\n",
    "                 'edgecolor':      'k'\n",
    "                }\n",
    "gs = fig.add_gridspec(2,1 + include_weights, wspace=0, height_ratios=[1,4])\n",
    "ax_legend = fig.add_subplot(gs[0, 0])\n",
    "ax_legend.set_axis_off()\n",
    "ax_legend.legend(handles[:n_accs], labels[:n_accs], loc='center left', title='Legend: accuracies', **legend_kwargs)\n",
    "if include_weights:\n",
    "    ax_legend = fig.add_subplot(gs[0, 1])\n",
    "    ax_legend.set_axis_off()\n",
    "    ax_legend.legend(handles[n_accs:], labels[n_accs:], loc='center', title='Legend: agent weights', **legend_kwargs)\n",
    "\n",
    "\n",
    "figname = '{3}__{2}__p_d_{1:.0e}__{0}'\\\n",
    "            .format(repr(hyperparams)[18:-1],\n",
    "                    sim_aggregator.aggregator_hyperparams.distance_penalty_factor,\n",
    "                    splitter,\n",
    "                    set_loader\n",
    "                   )\\\n",
    "            .replace(', ', '__')\n",
    "if not os.path.isdir('figures'):\n",
    "    os.mkdir('figures')\n",
    "fig.savefig(os.path.join('figures', figname + '.pdf'), bbox_inches='tight')\n",
    "fig.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Compare the model weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "FlatMapping({\n",
       "  'linear': FlatMapping({\n",
       "              'b': DeviceArray([ 0.74737775, -0.74737775], dtype=float32),\n",
       "              'w': DeviceArray([[-0.33786783, -0.24764758],\n",
       "                                [-0.76747066,  0.6531085 ],\n",
       "                                [-0.31852612,  0.3367445 ],\n",
       "                                [-0.08187868,  0.1588181 ],\n",
       "                                [ 0.5777908 , -0.25940347],\n",
       "                                [-0.7590264 ,  0.51881844],\n",
       "                                [-0.4437801 ,  0.4001845 ],\n",
       "                                [ 0.25706753, -0.16035254],\n",
       "                                [-0.5766699 , -0.17226446]], dtype=float32),\n",
       "            }),\n",
       "})"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_params['Weight Erosion (lr=3.0)'][0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "#features = ['fare', 'sex', 'embarked_0', 'embarked_1', 'travel_alone', 'pclass_0',\n",
    "#       'pclass_1', 'age_standardized', 'is_minor'] if set_loader == 'TITANIC_SURVIVED' else \n",
    "# ['quarantine', 'quarantine_missing', 'conjonctivitis_triage', 'diarrhea_triage', \n",
    "#       'referral(4-9)', 'pyrexia_triage', 'dysphagia_triage', 'hemorrhage_triage', 'myalgia_triage']\n",
    "features = ['Fare',\n",
    "            'Sex (1: female)',\n",
    "            'Cherbourg',\n",
    "            'Queenstown',\n",
    "            'Travelling alone',\n",
    "            'First Class',\n",
    "            'Second Class',\n",
    "            'Age (standardized)',\n",
    "            'Passenger is minor'] if set_loader == 'TITANIC_SURVIVED' else [\n",
    "                'Quarantine',\n",
    "                'Quarantine status unknown',\n",
    "                'Conjunctivitis',\n",
    "                'Diarrhea',\n",
    "                '4-9d referral time',\n",
    "                'Pyrexia',\n",
    "                'Dysphagia',\n",
    "                'Hemorrhage',\n",
    "                'Myalgia']\n",
    "features = ['Bias term'] + features\n",
    "fontsmall = 14\n",
    "fontlarge = 17\n",
    "def plot_weights(ax, params, title='Weights of the model', invert_yaxis=True):\n",
    "    \n",
    "    weights = np.concatenate((\n",
    "            np.array(params['linear']['b']).reshape((1,-1)),\n",
    "            np.array(params['linear']['w'])\n",
    "            ), axis=0\n",
    "    )\n",
    "    assert weights.shape[0] == len(features)\n",
    "\n",
    "    ax.barh(features, weights[:,0], align='center', color='k', zorder=1)\n",
    "    ax.barh(features, weights[:,1], align='center', color='gray', zorder=2)\n",
    "    ax.axvline(c='k', linewidth=3)\n",
    "    ax.set_title(title, fontsize=fontlarge)\n",
    "    ax.tick_params(axis='y', labelsize=fontsmall )\n",
    "    if invert_yaxis:\n",
    "        ax.invert_yaxis()\n",
    "    if set_loader == 'TITANIC_SURVIVED':\n",
    "        ax.set_xlim((-1., 1.))\n",
    "    else:\n",
    "        ax.set_xlim((-2., 1.5))\n",
    "    ax.grid(axis='x', color='k', zorder=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_weights_of(requesting_clients, aggregators):\n",
    "    if isinstance(requesting_clients, int):\n",
    "        requesting_clients = [requesting_clients]\n",
    "    if isinstance(aggregators, str):\n",
    "        aggregators = [aggregators]\n",
    "    \n",
    "    cli_agg_pairs = []\n",
    "    for agg in aggregators:\n",
    "        for k in model_params:\n",
    "            if k.lower().startswith(agg.lower()):\n",
    "                agg = k\n",
    "        for c in requesting_clients:\n",
    "            cli_agg_pairs.append((c, agg))\n",
    "            \n",
    "    ncols = max(len(requesting_clients), len(aggregators))\n",
    "    nrows=min(len(requesting_clients), len(aggregators))\n",
    "    fig, axes = plt.subplots(ncols  =ncols,\n",
    "                             nrows  =nrows,\n",
    "                             sharey =True,\n",
    "                             squeeze=False,\n",
    "#                              gridspec_kw={'top': 0.78},\n",
    "                             figsize=(7.5*(1 + (ncols > 1)),\n",
    "                                      5*nrows)\n",
    "                            )\n",
    "    for i, (c, agg) in enumerate(cli_agg_pairs):\n",
    "        agg_idx = i // len(requesting_clients)\n",
    "        c_idx   = i %  len(requesting_clients)\n",
    "        if len(requesting_clients) > len(aggregators):\n",
    "            ax = axes[agg_idx][c_idx] \n",
    "        else:\n",
    "            ax = axes[c_idx][agg_idx] \n",
    "        plot_weights(ax,\n",
    "                     model_params[agg][c],\n",
    "                     \"User: agent {c}\\n{a}\".format(c=c, a=agg),\n",
    "                     invert_yaxis=False\n",
    "                    )\n",
    "#     fig.suptitle(\"Weights of the Titanic survival model\", fontsize=fontlarge)\n",
    "    for axes_row in axes:\n",
    "        axes_row[0].invert_yaxis()\n",
    "    plt.show()\n",
    "    figname = 'Weights_{3}__{2}__agents_{4}__{5}__p_d_{1:.0e}__{0}'\\\n",
    "                .format(repr(hyperparams)[18:-1],\n",
    "                        sim_aggregator.aggregator_hyperparams.distance_penalty_factor,\n",
    "                        splitter,\n",
    "                        set_loader,\n",
    "                        requesting_clients,\n",
    "                        aggregators\n",
    "                       )\\\n",
    "                .replace(', ', '__')\n",
    "    fig.savefig(os.path.join('figures',\n",
    "                             figname + '.pdf'),\n",
    "                bbox_inches='tight'\n",
    "               )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA+sAAAFYCAYAAAArwW27AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAgAElEQVR4nOzdebgcRd328e8dQlgCsq8qBNlBEEmQHYICIhB8VHhEAQ0uQVmEF3lQFk2CCqLIooAYEMOmICI7ogQIm2whICAgQQg7WQhLEpJA4Pf+UTVJpzNzzpx1Jufcn+vq60xXV1dVz5mp6eqqrlZEYGZmZmZmZmbNo0+jC2BmZmZmZmZmC3Jj3czMzMzMzKzJuLFuZmZmZmZm1mTcWDczMzMzMzNrMm6sm5mZmZmZmTUZN9bNzMzMzMzMmowb62ZmZmZmZmZNxo11sy4iaYSkkPSRGtsvkBTdXa6eStKA/J5v3oZ9JOn/SZogaY6kpyV9T5K6sqxmjeS6qXu5bjKrn+un7uX6qfm5sW5mPcUAYDhQ9w8OcBJwOnAfcBjwIHAWcHxnF87Meq0BuG4ys+Y0ANdPTa1vowtgZt1HUv+ImNnocjQDSWsCxwIXRcTQHFy5Yn+CpFERMaVhBTTrRVw3zee6yay5uH6az/VT93PPulkTkbSLpNslvS5plqTnJV0qqX8p3n6S7pP0jqS3Jd0oabNSnNGS5kr6qKS/SnoTuDtvW1zSRpLWqKNMa0s6W9KTkmbm/MZI2q5K3BUk/UHSmzne1ZI+nIe0jSjFXU3SeZJelvSupGckHSepTyHOgLzviZK+JumpPOTqMUm7FeINBW7Pq5fkfSKH1/J5oB9wdin8HGApYJ/W3huz3sJ1k+sms2bl+sn1U0/mnnWzJiFpY+Am4Angp8B0YC1SxbcsMDPHOwb4JXA1cAmwDPBd4B5JgyLi6WKywN+BR4EfMv8C3YeBJ4GLgKGtFG0rYBfgr8DzwMrAN4Hbcn6P53L1Aa4DtgcuAB4BPgPcUOVYVyYNn1oSGAW8kvc7GVgb+E5pl/8BVgHOA2YBRwFXS1orIqYBd+Z9jwd+C/wz7/dPatsSmAs8XAp/CPggb/99C/ub9Qqum1w3mTUr10+un3q8iPDixUsXLMAIIICP1Nh+QfoKzls/MsdfpYU0Pwq8B/ysFL4aMA24rBA2Oqf36yrpDMjbRtdxHEtXCVsRmAycXwj7Qk7zhFLci3P4iELY74CpwBqluCeTKvsNSuV8C1itEG+LHH5YIWxwDjuwzv/PDcDLNbZNBq5u9GfIi5euWFw3zQt33eTFS5Mtrp/mhbt+8kJEeBi8WRN5M//9gqTFasT5EmlEzJ8krVxZgPeBe4FPV9nn3HJAREyMCMX8+41qioh3Kq8lLSVpJdJV5geAgYWonyNV+OWhUWcVVyQJ2I90Jfy90nH8nXRFe5dSGldFxKRCmR4B3gY+1lr5W7AUMKfGttl5u5m5bnLdZNa8XD+5furRPAzerHlcARxMunJ6qqSxpCuYl8f8iU02yH8fq5HGB1XCnu1IoST1I80UehDp6nTRc4XXawNTIuKtUpwJpfVVgBVyegfVyHbV0vrzVeK8QbpK3V6zgCVqbFsybzcz101FrpvMmovrp/lcP/VAbqybdZ3Z+W+tq4xLU6jUImK2pF1I9x99DtiNNNzrREnbRsRrzL9vam9qX9ksej8i3m1P4QvOAoaRJg+5h1TRfwAcB6zbjvQqx3AF6fiqKf9Ivl8jXkee6fkq8FlJi0XEvPTzD+xKpHvBzHoi103VuW4yazzXT9W5fuql3Fg36zoT89+NWPgKKcDGlK56Rrrp5+68nCDpc6QhT4cAI4FnctQXI+LRLihzNfsDF0fE94qBkk4qxXse2E3ScqUrxBuU4k0hDcPqFxFjOrGc0cb444FvAZ8ExhXCB5F+FMd3UrnMms3E/Nd104JcN5k13sT81/XTglw/9VK+Z92s64whXcE9VNICF8Yk7Uya6OPGQthKVdKozLa5fP57FWkWzpHFx3QU0lilnoK15fEjpCvBC+QlaUdgm1K8v5Gu1h5eCl/ghypfib0S2EfSVlXKtqykWkOsWlIZ7rZ8i7Hmu4404Uy5vIeSruxf344ymC0KXDclrpvMmo/rp8T1kwHuWTfrMhExVdKJpEeF3Cfpr6RZRzcjXZV8Hvh5YZcfSfo06V6riaTHihxMGsb055zmc5KOBU4HHpB0FfA66TElewCP0/rjRKBtjx+5Fvi6pBmkR4psnMv/b9JjUYrx7gV+Imkt4F+kx49UJjIpXr09jjQD6V2Sfk96PMqywKbAvqT3aGIdx1H0JGlo3KGSZpF+gO6PiOeqRY6IlyWdBlSeT3oHaXKWA4AfR8TkNuZvtkhw3eS6yaxZuX5y/WQljZ6O3ouXnr6QZiG9gzR86V3SxCJnA6uW4u1Cuvr7Iumq8iTSFdedqqS5F3BbTvMd0hCv0cA2hTijgbk1yjSA+h8/sixpVtRXc173A7vn9CeW4q5I+hF7K5ftL8D6Oa8flOKuBJxBusfqXdIjP+4BjgWWLJXzxCrlmlguP2mm1CdJV30DGNrKsfUBvp/fvzmkIXdHAWr058aLl65eXDe5bvLipVkX10+un7ykRflNNzPrEpI+SbqH6cCIuKzR5TEzA9dNZta8XD9Zhe9ZN7NOI6na7K1Hk+7duqObi2NmBrhuMrPm5frJWuJ71s2sM50haXnS/VeQhpztBvw2Il5qXLHMrJdz3WRmzcr1k9XkYfBm1mkkfZl0H9MGpGehTiTdh/XzKDyP08ysO7luMrNm5frJWuLGupmZmZmZmVmT8T3rZmZmZmZmZk3GjXUzMzMzMzOzJuPGulkvI2mApJA0tM74S0p6UdIRhbChOY0BXVTMbifpGEkTJfVrdFnMWiNp1/wd3KEd+/bN+57XFWVrNh15rzqxDGtIekfS5wphP5U0t1Fl6gqSzpN0Z6PLYYs+SYPz93ZwO/Yd2tnfeZ87dZykGyT9sdHlWNS4sW7WBbrih6KBDgf6ARd0d8aStpd0jaTnJc2SNEXSWEl7tzGdgyQ9Kml2TmukpMVL0c4DlgUO6bQDsF5D0hfzd/7gKtv+lLf9X5Vtv2nmkzdJK0oaUW9dJmm9fDy1ltFdXORmNRx4KiL+1t0ZS/ofSTdLejnXga9IulHStm1Io4+k70t6JqfxH0mHVYl6KrBd8aKEWYWklST9RNLDkt6WNCf/Jl8haUijy9fJGnbuBCDpQ5LOlvRaPn+6T9Jubdhfkv6fpAn5//S0pO9JUine2Fbq/O0L0U8G9pe0eWcdZ2/gR7eZWU2SFiM96/MPETGrAUXYABBwPvAasAzwReB6SYdFxLmtJSDpG8DvgZuA3wCfAH4ErAl8uxIvImZIugQ4RtI5EfFBZx+M9Wh35b87An8obdsBmJu3/bLKtpciYmIb87sNWAqY08b92mpFUkNzNnB3G/a7ArihSvh/O6NQbdRd71VVklYGDgaqNW67w+bAm6T6bwqwEnAgcJekIXVeQPgZ8EPgYtJn/TPA2ZKWiYhTK5Ei4jlJNwPHAt1+YcKal6RPkn6HVwSuJP0uvwOsBewJXCdpWESc37hSdo5GnzvlBvX1wKeA04EXSHXQTZJ2jYh6nt1+EnAicCnpItwuwFmkTo2fFeL9jOoXJH5Dek78g5WAiPinpEeB/wMOauNh9VpurJtZS/YE1gAub8/OkvpHxMz2Zh4Rf6DU8JH0G+AhUmXfYmNd0pKkH5mxwN6RH38h6U3geEm/jojHCrtcDhwJfBoY095yW+8TEVMkPUVqfM+Te8w/Qjrh2UuSCp/DDwGbAX9uR34fkBrQzeqRiLi0rTt1tM6opgneq4NIFx2vas/OkpaOiHfam3lEnFQlzXNIj4c6mlYa1ZI+Qnqs1O8j4ls5+ILcIPixpAsi4vXCLpcDF0taJyKea2+5reeQtBxwHel7MDAiHi9FGSFpV9Jjy3qChp47kTo1dgIOjojROc2LgMeBXwGDWsl/TdIFt4siYmgOvkBSACdIGhURUwAi4pYq+w8ElgfOjYh3S5svJ9Ubh0XE2+08vl7Fw+DNGkjpnqaTle6Vfjf/PVnSElXi7iLpH5LelDRT0mOSjits30zShXmY4ixJ0yRdK2mTDhTxC8DkiHikjmMZm/P+uKRbJE0HLutA3lXlZ46+RPohaM0uwMrAOZUGUnYu6aThf0vx7wfeIv3QmbXVXcD6klYrhO0IvAucCawAbFrYth2wGPN75QGQNEjp3r4383f5gfIQUdW4D1vSlyQ9Xhi2+A21cG+0pL0k/SsPbZ4gaf9iHsCEvHpKYVjjiW16V2qQdGnOd21JV+eLaGML23eUdLukGZKm53pl61IafSWdoDQse1Z+zx6WNKx4HDXeq3rS/1bed0dJp2r+kNK/q/5bF74APBgRb9TxnrykNGR9F0n3SppF6uHqVLkh8Dr11aNfABYHzi6Fn0NqXJVvS/oHqX79QgeLaT3HIaSLlkdXaagDEBFjIuK61hJSuj3u1vydnZFf17qlYwlJZ0ialM+bbpK0Xim9nnju9L+k0TTzLphGxGzSaIaBktZtZf/Pk4bwV/vOLwXs08r+lV7zi6ts+0dOY49W0rDMPetmDSJJwF+Bz5Eq1HtJJ+/HkXrbhhTifhW4BHgWOAOYBGxEqlBPydF2BzbJab1M+mE8hDTUcdOIeK0dxdye1Itdrw8Bt5CuoF8JzMrlX5r6r5hPKw9Bl7QssASpsfN5UiX/1zrS2jL/faAYGBGvSHqpsL0SHpLGkRpYZm11F+nWih2Y34u6IzAOGA+8kdcrJ6s7FPYD0qRKwM3AY8BPSUO3vwxcK2n/iKjZCy9pL9L37ingeNJ37hTg1Rq7bAP8D2m+hqnAMOAySQ9FxIRczmOA01hwWHurJ6DA0krDv8umR0RxOHof0snbQ8APSI08JO0C/B14kfQ+9AG+A9whaXBE3Jf3H0mqM39P+p4vBXyc9N6OqlW4NqRfcQapPjsZWIX0vlxCK3WF0oSVW5Fu5anXeqT67QLgQtItQEhaBliyjv2j1NNdKcvypEb3KqQhsRsy//ejJVuSPoePlsIfBCJvv6iQ+WRJz5Pem9PrSN96vn1I35+/dCQRSTuRzjFeIX0XIZ3n3C7pMxFxT2mXX5FuQTqF9Lk/EhgrafOImJbj9MRzpy2BhyOifJH2gcL2lm5J2pL0vj1cCn+INLR9S1KduxBJfYGvAE9HxP1VojxKGum0I+0YVdYrRYQXL146eQGGkk5idmghzt45ziml8F/m8D3z+rKkK6SPAcuU4qrweukqeaxHqhRPKIQNyOkPbeUYFiNVyme3cHwDCmFjc9jRVeKPyNvqWQZU2f8vhe1z8/qKdfwfzs77LF5l2wOkH7Ny+Kich1pL34uX4gKsnT9vZxbCngROza9vAP5Y2HYHqXdTeb0P8DRwO9CnEK8P6WLexELYruU6htS4fg1YrhC2Sf48zy2E9c37vgdsXAhfkzQK4JRC2Ho57g/rfA8q8WstQwtxL81hp1dJ55H83qxSCPswMB34ZyHsMeDaVspU7b2qN/1v5X3vKP1PjsnhG7aS9wY53lFVtv20+H/JYS/l+PtUiV95v1pb5tYoyyOFOLOB3wJL1PE/vRl4vsa2acCVVcJvA/7T6O+kl+ZY8ufkkSrhy5BGv1WWDxW2Dc6f1cGFsHGki56rFcLWII2Ie6AQNjTv+xSwVCF8N0rnXfTAcydgBnBZlfQ2yXGPbOUYbgBerrFtMnB1C/vulfM4sYU4TwO3Nvpzuags7lk3a5zK0MHTSuG/IJ0I7k2ajGV3YDng0IiYUYwYudbLr+fd0yipP6kH5k1SpTiwHeVbidTL1erQzYIPSL10ZRdT/+RU1a5ij8zprknqWe9Luto8rUrcoqVIb9N7VbbNJl3NLptG+rFdnrYdu/VyEfG8pBfJPea5Z3lD5n/27wG+m7dVelxvKXyPtwTWJ82zsKIWnHT3b8BISetGxEI9IpLWIg2x/2VEvFUo0xOSbiGdpJbdHhFPFuK+ImkC8LE2H/zCfk/1+zX/XSVsgbknlO6R/gRwRuT7InP5XlZ67M8wSStF6j1+E9hM0kYR8VQ9BWtj+hW/iwVH/FQmaPoY8J8WsquMLmhLXfJSVB8OfAowuo79a02OeQjpt+SjpJ6vJfPS2sR7LU3ONztvL5tGeo/NIP3WTq8SfibwzcL6raSLawuRtDrpXOY3ETGpEh4Rr0q6FDhU0mrFbaTv7axC3FskPUE6vzouh/XEc6da39nZhe0tac93vuJrpMb6JS3Emcb8utFa4ca6WeMMAKaUTgiJNFHVVGCdHFS5v6o4EdpClCZwORnYjzTcq2hqB8qp1qPM81pUmQgpIp4lDeFvl0iTwFWO/+J8Qn2bpM1iwSG1ZbNIdxwsXqXBvmTeXtaW4zUruwv4ch6yXBnmfk/h78mS1ib14i7Fgverb5D/XkDtx/2sSvXhi2vnvxOqbJtA9cb681XC3iDN1txRz0REPZM0fkCa6KxoQP5brfH9RCHO68AJwNXAk5L+QxpKemVEtPSs77akX1F+ryon4vW+V22pV6rWlRHxb6pf7KhLFIakKj1C727ScPvPtLLrLNJtSNW4HrV6vE0aJVj2K+Zf1GttQsoB+W9r39tiY73ahbT/UKgPe+i5U63v7JKF7e3Zv5JG1f2VJk3dB7gzIqr9vsyL2kr+VuDGullzEunKZOU1hfVaLicNGzuddH/sdNKJ8Jm0bzLJ13OeK7Rhn1oV+DKk4W71mBJpErmWXE7qGarcv1ZL5V7dNVn4ZHsNFr4HE9LxziVdWTdrq7uArwLbku7JezLm3xv5AGmY+Y6kz2QlfkXle/pDat/v+ESN8JbUOjGq9T3rzhOp92Lh+ypbskB9GBF3SvoYqadsV+BLwOFKj188vB3lqVXftve9qpzsd0Y9uhyt94hBGk00qZUI70u6EviVWp+1/VVgsKQ+xdEFSk/bWJ50/3DZCnSsoWM9y5PAlpL6RWF28Dyy50kASR15YkOt722186byd7Ynnju9yvzfmKI18t9q39miV4HPSlqseD6WR4St1ML+/0tqzFebWK5oBdJ8IVYHN9bNGmciqTJcYLhlHjq7EvN7myo9ZZszf2KqBeSJg/YARkTEyNK2dp005ZO5CUBrs4bW4xjSs5rrsQ4L97SVVU5YW/sxHJ//bkWhsa70WJKPUH1I6brAU8VbDMzaoNKju2Ne5g1hjIjZksYzv7H+Dgs2yp/Jf6fX2StdVPl8r19lW7WwejXiezAx/92oyrZK2Lzvcx72fxlpcrzFScMvD5N0SkS83NH0O2giadhoZ9Sj5wAH1BHvfeo7vyvWoy011seT7rXdnAUnF/wUqeEzvso+69YIt97petIEuvvR/pnOJ+a/bfnebsTCjybcoJJWDz53Gg/sLqlv6WJo5WkX5YnjysaT5uv4JGmegIpBpAsYtb7bB9HKRIK5jv4oaYJPq4Mf3WbWONfnv0eXwv+vtP0fpMlTjstXWefR/JtaK1c++5S2H0j1q6v1uodWnsdZp4tJw87qWebddyVp1XJCkvqQZq0OCj8ikpaTtFHufaq4nXSV+7DCewVwaP57ZSltke5Rq/ceMbOyJ0mfuc+STnTKn6V7SCNCtgPuLZ1IPUga8nhM6XMMgKTyEM15IuIFUq/7QcV98+OHqt4DWqfKs37recRXp4iIl0iNwq8VZ5TPF9kOIE0A93oOW6m073vMv6hZtcxtSb8TjuVd0v91q05I7hTqq0N3L+5Uox7tTzqxnkHu2czhK+d6tNiDfy1ptFF5pMKhpBPzG4qB+XO6Fq5Hbb7zSL2xp0vatEacFkepRJqVfRypjpv3mc73sh9EmmCuPKJkWPGzLGk30iRrN+agHnnuRGosLw8cWAnII2G+QZpY95lCeOU7X5x1/jrSBKTVvvOzmX9+Oo/Soyx3BK6Jlp+fvhmp9931Q53cs27Wtb6WH8VUdilp8ribgePzhEf3kx6ldBBwQ0T8DSAipks6gvRonIclXUKqlDcgnfBvl+PcDhybK+T/kn4o9qUD94oD1wAHSxoYEW15DMkCOnDf1c2SXgXuI/3Qrw7sT3o00+k53YovAH8gPZJodM53ttKz6EcB10u6hjTp0aHAhRFRHga/NWkCpqvbUVYzIiIk3U2aCBGqN9a/n18v8Hz13CNzMKleeELShaSeotVJdcP6pAnrajmO9J29V9LvSZMwHk6a7+Hj7Tye1yS9DBwoaSLp9pBHI6K14fhb5BPesikRUU+PytGknpf7JJ1POpH/LunRY98vxJsg6TbSCIXJpPfncOBftHzLQL3pd4ZrgFMlrRwR7R4a3oF71p/Kkww+AkwhNaS/nv8eWpyACziKNA/AvFEhEfGCpDOA/8uPZbqbdJ/7l4Hjq1zY+CzpYuo17Sir9UAR8aakz5MayePzLRj3kS72rEm6z3ktoLURRd8n3fp2n6Tfkb63h5Aaf+WOD0iTpN2Vz5tWJT267VXSU3fowedOV5G+p79Veq78i6TRMQNYeP6Sw0m997uQZqavTLZ5GqmTqA9pQs1dSBczfxwRk6vkeSDp/9HaEPjdSf/3m9t6UL1WV0wx78VLb1+Y/3iOWsuuOd5SpN6S50n3sj5PmuhkySpp7kZ6HM50Um/XY8APCttXB/5IGrY1I8cdSKp8xxbiDaCOx4/kuH1JjeRf1ji+AYWwsaRJpTrzfTw0pzuZdJX3DVJv+VdaeM8XOi7SienjpB/uF4GTqP44tzOAFyg8osmLl7YupBPKoMqjb0gnjJV64NM19t+M9PzZyYXP7PXA/xbiLPQ4shy+H6mROoc0kdLXSPdeTi/EqTy67bwqed8NjCmFfYY0bHIOrT+Sp7VHt91XiHspMLuFtHbK3/eZuU4bA2xTinMi6bF2r5N6fP6bv8cr1/Fe1ZN+5dFt5fDKcR5Yx+dhlfzeHVIKr/Xotps7+fN4Euli8Ou5Hp1M6g3fvUrcn9Z4r/qQRn39Nx/LBOCIGvldT+E3x4uXykKaAfynpItpM/Jn6YVc3+1TijuY0qPbcvgOpPObGcw/19muFGdo3vczuf6bTLrt6GZg/VLcHnfulNNdjvSkjUmkxvEDwGerxBtR433uQ/ote6bwnT+KGo+1JU389yqwWCvlehi4tNGfxUVpqTzb1cysKknHAMcCa8eCPTA9Sh4W+gIwMiJ+3ejymHUWSTcCH4uIjRtdlt5K0nnA1hHxyUaXpStJWof0yKvPR8RNjS6PWaP0lnOntpC0DfBPYMuIeKS1+Jb4nnUza83ZpKuq3250QbrYd0ijFqo969Ss6UlaPA9TLoZtShqWfFtjSmXZSGBDSXs2uiBd7FjS6Ak31K236y3nTm1xAnCFG+pt4551MzOzHiBP8HMraXj5S6TZiL+bN28RLT+ey8zMzJqMJ5gzMzPrGd4gzTz+TdJ90rNIj5I73g11MzOzRY971s3MzMzMzMyajO9ZNzMzMzMzM2syHgZvnapfv36x+eabNyTvCRMmsP766/eqvBuV70MPzX9s6MCBA7s9f/+vu9dDDz00NSJWaUjmncR1U+/Ju5H1U298vxudt+unjumNv2m9sW6C3vme94S6ycPgrVP1798/Zs6c2ZC8Bw0axLhx43pV3o3KV9K8142oQ/y/7l6SHoqIQQ3JvJO4buo9eTeyfuqN73ej83b91DG98TetN9ZN0Dvf855QN3kYvJmZmZmZmVmTcWPdzMzMzMzMrMm4sW5mZmZmZmbWZNxYNzMzMzMzM2sybqybmZmZmZmZNRk31s3MzMzMzMyajBvrZmZmZmZmZk3GjXUzMzMzMzOzJuPGupmZmZmZmVmTcWPdzMzMzMzMrMn0bXQBzBpp5MiRnZLOkCFDOi2tRSHfIkm9Kt/OyDsiOqkktqho1Pe0kXVEb62fFuW6qb0GDhzYkHzNGqkj9Vtbvqs+Z+jd3LNuZmZmZmZm1mTcWDczMzMzMzNrMm6sm5mZmZmZmTUZN9bNzMzMzMzMmowb651MUkjat9HlMDMzMzMzs0WXG+t1kjQ6N8Qry1RJN0jaqBR1DeD6Li7HDV2VvpmZmZmZmTWeG+ttM4bUGF8D2B1YCri6GCEiXouIOQ0oW5tI6qtGPl/GzMzMzMzManJjvW3m5Mb4axExHjgD2EjSUpUI5WHwkn4u6T+SZkmaKOkXkpYsbP+opGslTZP0jqSnJO1fLXNJI4CvA3sVevgH520flnS5pDfycqOk9Yv7Snpc0lBJ/wXmAP0ljZX0W0m/ymWYIulISUtIOkfSm5JekHRQ576VZmZmZmZmVkvfRhdgUSVpWeDLwGMRMauFqDOBbwAvA5sA55Eayj/K288FlgR2Ad4GNmwhrdOAjYEVgUrjeZqkpYHbgX8COwPvAscAYyRtHBHv5LjrAF8F9stxZufwA4DTga2BfYAzgT2Am4FBpAsEF0i6NSJeaaF8ZmZmZmZm1gncWG+bPSTNyK/7Ay8Ce7a0Q0T8pLA6UdLJpIZ0pbG+NnBVRPwrrz/XQlozJM0i9/BXwiUdCAg4OCIihx0CTAb2Bv6co/YDDoqISYV9Af4dESPy+unAD4H3IuKsHHYS8ANgO+Av5XJJGgYMA+jXr19Lb4eZWbdx3WRmzcr1k5nVw8Pg2+ZOYIu8bA3cBvxD0kdr7SBpX0l3S3otN/TPANYqRDkLOFHSvZJ+KmlgO8o1kNRrPl3SjJzPW8AKwLqFeC8VG+oFj1Ze5Mb+ZOCxQth7wBvAqtUyj4hRETEoIgb17evrP2bWHFw3mVmzcv1kZvVwY71t3omIZ/LyAPBN4EPkK6NlkrYBLgf+DgwBPgmcCCxeiRMRvyc1tP8AbAD8M9+b3hZ9gEeYfyGhsmwA/K4Qb2aN/d8rrUeNMH9ezMzMzMzMuoEv5XVMAB8AS9fYvj3wcnEovKS1F0ok4iVgFDBK0g+AI4ERNdJ8F1isFDYe+AowNSLebMsBmJmZmZmZWfNxT2nbLCFp9bxsDPwGWIbaz1V/GviwpAMkfUzSd0mN6nkknSVpj7x9C9LEbk+0UIaJwMclbShpZUmLA5cBk4BrJe0saR1JO+UZ3tdvIS0zMxZ5fYkAACAASURBVDMzMzNrQm6st82uwKt5uR/YCtgvIsZWixwR1wO/JM2u/iiwG/DjUrQ+pEb/E8AtpEb311sow/nAk8A4YAqwfZ7tfSfgWeBK4CngItI962+08RjNzMzMzMyswTwMvk4RMRQYWkc8ldaPA44rRfttYfsRbSzHFGD3KuGTgINb2G8EVYbWR8TgKmEfrxK2elvKaWZmZmZmZu3nnnUzMzMzMzOzJuPGupmZmZmZmVmTcWPdzMzMzMzMrMm4sW5mZmZmZmbWZDzBnPVqw4cP75R0Bg0axLhx4zolrUUh3xEjRsx7HRHdnn+jjrvReduiq7Pqmrbqjd+VRtZPvfH9ruRt1tu0tV5v9LmTLZrcs25mZmZmZmbWZNxYNzMzMzMzM2sybqybmZmZmZmZNRk31s3MzMzMzMyajCeYM2tykhZYL05QMmTIEEaOHNnNJTKz3qpcH7VkxIgRrp/MrCFaqnt87mSLEvesm5mZmZmZmTUZN9bNzMzMzMzMmowb62ZmZmZmZmZNxo11MzMzMzMzsybjxvoiTNJoSVFl2aLRZTMzMzMzM7P282zwi74xwEGlsKntSUjS4hHxXseLZGZmZmZmZh3hnvVF35yIeK20zJW0p6S7Jb0paZqkv0nasLKTpPVyL/yXJd0uaTbwzbxtB0l3SZol6SVJ50hatmFHaGZmZmZm1su4sd5z9QdOB7YCdgHeAa6XtHgp3s+B3wAb5+1bAH8HrgI2B/YFBgHnd1O5zczMzMzMej0Pg1/07SFpRmH9roj4XERcWYwk6WDgTWAgcF9h05kR8ddCvF8Cl0bEmTlogqTDgAclHRoR07rmMMzMzMzMzKzCjfVF353AsML6LABJ6wMnAVsDK5NGUQhYiwUb6+NK6Q0EBkg6oBCm/HddYKHGuqRhlTL069evvcdhZtapXDeZWbNy/WRm9XBjfdH3TkQ8UyX8RuA54NvAK8AHwBNA+RdhZmm9D/A74NdV0nypWgEiYhQwCqB///5Rd8nNzLqQ6yYza1aun8ysHm6s90CSVgPWB74ZEXflsE9R3xwF44FNa1wAMDMzMzMzs27gCeZ6pqmk4erD8qzvg4FzSL3rrTkF2C7PAL9F3n+IpPO6sLxmZmZmZmZW4MZ6DxQR7wNfBrYEHifN9n4c0Ooz1CPiEWBnUs/8XcAjwM+AV7uqvGZmZmZmZrYgD4NfhEXE0Ba2jQE2LQUvWdj+DPMnjivv+wCweycU0czMzMzMzNrBPetmZmZmZmZmTcaNdTMzMzMzM7Mm48a6mZmZmZmZWZNxY93MzMzMzMysyXiCObMmFxE1tw0aNIhx48Z1Y2mSESNGdHueZtZ4LdVHZa6fzKxRhg8fXnOb6yZblLhn3czMzMzMzKzJuLFuZmZmZmZm1mTcWDczMzMzMzNrMm6sm5mZmZmZmTUZtWWyGLPW9O/fP2bOnNmQvNs6YYikLixN7+EJU1qeyKazNGpCHABJD0XEoIZk3kkaUTe5jmk810/t09Y6zfVTxyxK506NzLcn1anNWjd11fnMovQ56yydVTe5Z93MzMzMzMysybixbmZmZmZmZtZk3Fg3MzMzMzMzazJurJuZmZmZmZk1GTfWM0m3S/paF6S7tKS/SHpLUkga0Nl5tLE8EyUd04b4e0t6RJI/K2ZmZmZmZt2kaRpgklaRdG5uTM6RNEnSrZJ264a89wI+ClxWCBuWG/BvdrCR/Q1gJ2AHYA3gxQ4Wt1tFxA3A+8ABjS6LmZmZmZlZb9E0jXXgKuBTwDeBDYC9gb8BK3VD3kcCoyPi/ULY0sA/gBEdTHs94MmIeCwiXivlsaj4A/C9RhfCzMzMzMyst2iKxrqk5YEdgR9GxK0R8XxEPBgRp0XE5YV4/SSdKuklSTMlPSjps4XtP5L0mqRVC2F/kjReUr8aea8C7ApcVwyPiDMj4hTg7g4c11jShYCdcu/82DqPY3CO/zlJD0maJekuSR+RtLOkf0maIekGSSsV9ttK0j8kTZX0tqS7JW3bShmXkzRK0mRJ0yXdIan8TMDrgEGS1mvve2FmZmZmZmb1a4rGOjAjL/tIWrKFeH8Adga+CmwGXARcL+kTefvJwATgQoB8D/rnga9GxLs10twBmAP8u62FljS20gCv4Yu5zPeShsB/sc7jqBgJHAVsDawAXAH8GBgGDAY2ZcGe/2WBS0gXPj4FPALcJGnlGuUXcCPwYdJIhk8CdwK3SVqjEi8iXgAm5TKbmZmZmZlZF+vb6AIARMRcSUOB84Fhkh4G7gGujIj7ASStC3wFGJAbjwBnS9oVOAQ4NCLel3Qg8C9JvwC+A3w/Ip5qIfu1gcntHJ7+QksbI2KapHeAdyPitXqPo5DEjyLirrzfecBvgIERMT6HXQTsW8jvtmL+ko4AvgTsAVxapYi7AFsAq0TErEqekoYABwG/KMR9BRhQ7TglDSNdQKBfv6oDGMzMup3rJjNrVq6fzKweTdFYB4iIqyTdSOoV3pbUwPy+pBMi4mRgS0DAE6lDeJ4lgNsK6Twv6UhgNHBjRPy2layXAma3s8ztmT2+ruPIHi28npT/PlYKKw75XxX4CakRvhqwGOn41qpRloGke/OnlMqyJLBuKe6snNZCImIUMAqgf//+USMvM7Nu5brJzJqV6yczq0fTNNYBImI2cEteTpJ0ATBC0mmkIfsBbAW8V9p1Vml9J9IM5mtJWiIi5rSQ7VTSEPPu0pbjKG4PgIgohxVvZbiI1Ej/f8BE0vD+W4Fal2z7kBr8O1bZ9nZpfUVgSo10zMzMzMzMrBM1VWO9iidIZVwSeJjUI716RNxeawdJXyQ9ZuzTwMXAKcDRLeTxMLCKpJUjYmpnFbyV/Fo9jnbaAfheRNwIIGk10r3ytYwnNe4/iIhna0XK8wism+ObmZmZmZlZF2uKCeYkrSTpNkkHStpc0jqS9gOOBW6NiLcj4mnSc9BHS9pX0sckDZJ0TG6gI2lN0n3vx0fEncCBwBGtPKv9YWAyqaFbLNPqkrYgPUYOYBNJW0hasRDnYkkXt+VY6zmODngaOFDSJpK2Ai4Hak2sBzCGNDfAtXnm+XUkbStppKRib/s2pF76ezpYPjMzMzMzM6tDUzTWSTPB30d6zNkdpJnZTwb+CHy5EO9g0kzqvwCeAm4gDXl/Ps9sfhGp8X0GQETcDfyc1DCu+rz2PLHchaTe+KLv5LQuy+s35vV9CnHWovb94C2peRztSKvoG8AywEOkhvqFpOHwVUVEAHuS7pU/H/gP8GdgQ9KEchVfAS6LiHc6WD4zMzMzMzOrQ1MMg8/3lB+fl5bivUd6VNmIGlEW6kGPiB8BP2qlCGeQJnz7WGU4eES0lE8l7cGtpEtEHF4lrMXjiIixpKHyxbC/VAk7DzivsP4v0mPeii4p7TOgtD6ddJHkyGplyc+h3xcoP3vdzMzMzMzMukiz9Kw3VERMJvVKf7TRZWlC65Aei/dcowtiZmZmZmbWWzRFz3oziIjrGl2GZhQRDwAPNLocZmZmZmZmvYl71s3MzMzMzMyajHvWrddK8+t1jkGDBjFu3LhOS6/Z803zOSbDhw/v9vwbddyNztsWLZU6prd+Xntj/dQb32+z7tJZ5229sW4C1xGLKvesm5mZmZmZmTUZN9bNzMzMzMzMmowb62ZmZmZmZmZNxo11MzMzMzMzsybjCebMullxgpFmTK+tRo4c2e15DhkypNPzbcRkL2aLqnrrnd5WP7WlbnKdY9Z7VeoJ1wPWGvesm5mZmZmZmTUZN9bNzMzMzMzMmowb62ZmZmZmZmZNxo11MzMzMzMzsybjxno7SQpJ+3ZR2mMlnd0VaZuZmZmZmVnzc2O9BkmrSTpL0n8lzZH0sqS/Sdqz0WUzMzMzMzOzns2PbqtC0gDgHmA6cBzwL9KFjc8A5wFrdVG+fYH3uyLtOvPvAygiGlYGMzMzMzMzc896LecCAgZFxJ8j4j8R8WREnA18ohBvRUlXSpop6VlJBxYTkfRhSZdLeiMvN0pav7B9hKTHJQ2V9F9gDtA/b+6be/Yr+/4yN6Yr+64g6aK8bZakMZI2LWwfKmlGqTyD8/D9lYtxJO0p6XHgXWBjSX0lnVHI+wxJv5U0tjPeXDMzMzMzM2uZG+slklYE9gDOjogZ5e0R8UZh9cfAtaQG/BXAhZLWzuksDdwOzAZ2BrYFXgXG5G0V6wBfBfbL6czO4QeQ/j/bAocAw4CjCvuNBrYGPg98CngHuFnSUm085CWBE3MemwDPA8cAQ4FvAdvkcny1jemamZmZmZlZO3kY/MLWI/WqP1lH3Esi4lIAST8CjgR2JDV498/pHBwRkeMcAkwG9gb+nNPoBxwUEZMqiUqC1LD/Xt73KUkbAEcDp+fe+X2AnSPizrzPQcALpEb+BW043sWAIyLioUL+RwKnRsRVef0o4LO1EpA0jHQxgX79+rUhazOzruO6ycyalesnM6uHe9YXpjbEfbTyIiLmAlOAVXPQQFKv+fQ81HwG8BawArBuIY2Xig31gvsqjfzsXuDDkj4EbAx8kMMq+b8FPEbqHW+LucAjlRVJywGrAw8U0g7gwVoJRMSoiBgUEYP69vX1HzNrDq6bzKxZuX4ys3q4dljYBCBIDeKrW4n7Xmk9mH8BpA+pEbx/lf2mFV7PbEcZW7qgUGngf1Al3uJV4s+pMaFcVAkzMzMzMzOzbuCe9ZKImAb8HThc0jLl7ZKWrzOp8aQh9VMj4pnSMq21nYGtlcfDZ9sAr0TE28ATzL+fvVKuDwGb5W2QevmXzuEVW7SWae6hf410H3wlbQFb1VFmMzMzMzMz6wRurFd3KKlXepyk/SRtKGkjSd+lMPS9FZcBk4BrJe0saR1JO0n6VXFG+BasCZyZ894X+D/gDICImECa2O53knaUtBlwKfA28Me8//2kXvtTJK0n6Uv5uOpxFnCspC9I2hD4FbAG7m03MzMzMzPrFm6sVxERzwFbArcAp5Ia6LeRJnU7pM403gF2Ap4FrgSeAi4i3bP+Rgu7VlxGmvztfuB84Pfkxnp2MOm+8uvy36WBPSJiVs5/Gmmyud1I97IPA35UT9mB04BLgD8A9+Wwq5k/U72ZmZmZmZl1Id+zXkNEvAockZdq2xe6bzwiBpTWJ5Ea1bXyGAGMqBI+uLB6eI193wC+XivtHOdaUg980aWF7aNJj4Ar7zeX9Ji4eY+KkzQeuKel/MzMzMzMzKxzuLFuC8nPiv8scAfpMzKM9Az4YY0sl5mZmZmZWW/hxrpV8wHwNeCXpFslngA+FxHjGloqMzMzMzOzXsKNdVtIRLwI7NDocpiZmZmZmfVWbqybdbOIzptUf9CgQYwb1/0DHopPFRw+fHi359+o4zazpJ56rDfWT66bzKwejTh3skWTZ4M3MzMzMzMzazJurJuZmZmZmZk1GTfWzczMzMzMzJqMG+tmZmZmZmZmTcYTzJk1ueJkSe3Z3tVGjhzZ7XkOGTKkw/l6chezRUt76rrOqJ9cV5g1Vlec5/jcqTnydv3aOvesm5mZmZmZmTUZN9bNzMzMzMzMmowb62ZmZmZmZmZNxo11MzMzMzMzsybjxrqZmZmZmZlZk+nxjXVJH5Y0StJLkt6V9LKk8yV9pNFlaytJgyWFpJUbXRYzMzMzMzPrOj26sS5pHWAc8HHg68B6wIHApsCDkgY0rHBmZmZmZmZmNfToxjpwDvABsGtE3BoRL0TE7cCuOfwcAEljJZ1d3FHSaEk3FNYl6VhJ/5U0S9Jjkg4s7fNhSZdLeiMvN0pav7B9hKTHJe2f05ku6ZpiT7mkzSTdKuntvP1fknbJFxZuz9Gm5B720XmfJSSdKWmSpNmS7pO0QyHN+yX9oLB+Wd5/9by+dB51sH3h/ThX0smSpkqaLOk0ST3982JmZmZmZtYUemzjS9KKwB7AORHxTnFbXj8X+JykFepM8qfAN4HDgE2AU4DfSdor57c0qTE9G9gZ2BZ4FRiTt1UMAL4MfAHYHfgk8LPC9j/m/T6Vt43Iab4IfCnH2RRYAzgyr/8ip/mNvM9jwM2S1sjbxwK7FPLYGZgKDM7r2wPvAQ8U4hwAzAW2Aw4Hjsp5mJmZmZmZWRfr2+gCdKH1AQFP1tj+RN6+fo3t80jqDxwN7B4Rd+Xg5yR9itR4vxHYP6d3cERE3u8QYDKwN/DnvF9fYGhEvJXjjAIOLmS3NnBaRDyV158plGNafjk5IqYWyvZd4FsRcWMO+w7w6Vy2E0mN9cMk9QXWAZYDfk1qwF9OarT/MyLeK74/EfHj/PppSd8GPgP8qcr7MwwYBtCvX7+a76OZWXdy3WRmzcr1k5nVoyc31iuiRrjy33frSGMTYElSb3UxvcWBifn1QFJDeLqk4r5LA+sW1p+vNNSzV4BVC+unAxdI+jpwK3BVoeFezbq5HPdUAiLifUn35nID3AUsAWxFun//LmAMcF7ePhi4qZTuo6X1cjnniYhRwCiA/v3713q/zcy6lesmM2tWrp/MrB49ubE+gdRQ3xS4psr2jUnDvJ8j3b+u0vbFC68rtwsMAV4oxXuvEOcRUg972bTC6/dK26KQPhExQtJlwOeAzwLDJX0nIi6ski6Fcler6COnOUPSeFJP+qak4fr3Amvne+q3Ao6tcVxVy2lmZmZmZmZdp8c2viJiGnAzcGjpnvHK/eWHAVfnXu4ppHvAiz5ReP0EMAdYOyKeKS3P5zjjSbPNT60SZxptEBETIuLXEbEX8HvgW3lTZRTAYoXoz+Tw4oRyi5HumX+iEG8sqbG+MzA2ImYD9wMnsPD96mZmZmZmZtZAPbaxnh1GatiOkfRpSR+VNBi4hdRA/V6Odxtpsrl9JG0o6XTgo5VEImI6cBpwmqRvSFpP0haSvpPvOQK4DJgEXCtpZ0nrSNpJ0q+KM8K3RNJSks7Jz1MfIGlrUiO80uh+ntTDvZekVSQtExEzgd8CP5e0p6SN8/pqpEn0KsaSGuvLki4sVMIOZOH71c3MzMzMzKyBenRjPSKeAwYB/wYuId1ffjtp2PsWEfFajnphYbkHmAFcXUruR6SZ2Y/J6d1Cmp39uZzXO8BOwLPAlcBTwEXACsAbdRb5/Rz/IuA/uQz3kia3IyJeBoaTZo+fBFQeN/cD0gR2fyANxd8c2CMiXi2kfRepoX9XRLyfw24nXcwYW2f5zMzMzMzMrBv05HvWAYiIl4BvV9YlHQH8itSwvjbHeY/UC39YC+kE8Ju81IoziQVndi9vH0Fq8BfDRgOj8+t3ga+2cjw/AX5SCptDerTaUS3sN4MF78MnIsay8L36RMTgKmFDWyqXmZmZmZmZdZ4e3bNeTUT8hjT0exNJSzW6PGZmZmZmZmZlPb5nvZqI+HPrsczMzMzMzMwao9f1rJuZmZmZmZk1u17Zs262KEnTJVQ3aNAgxo0b142lSaT5Ux0MHz682/Nv1HGbWeO0VBcWNbp+MrPOVe93v14+d+p+Pm9rP/esm5mZmZmZmTUZN9bNzMzMzMzMmowb62ZmZmZmZmZNxo11MzMzMzMzsybjCebMGmTkyJEdTmPIkCGdkk5HFCdM6Q35dkfenT2Zjll3GzlyZK+tn5q9bnL9YtZ4jagb21Mne5LOxnPPupmZmZmZmVmTcWPdzMzMzMzMrMm4sW5mZmZmZmbWZNxYNzMzMzMzM2sybqxXIek+SafVu95Ikl6TdHijy2FmZmZmZmadp9tng5fU2jSkF0XE0O4oSwfsCbzX6EKYmZmZmZlZz9SIR7etUXi9N3B+KWxWtZ0kLR4RTdFAjohpjS6DmZmZmZmZ9VzdPgw+Il6rLMCb5bCIeEvSRpJC0n6S7pA0G/i6pNUkXSHpZUnvSHpc0gGVtCUdKelFlR40Kumvkq4orH9R0sOSZkt6VtIISYvXewxVhsW/JukHki6UND2X4XulfTaRdE/O8wlJu0maK2n/FvLZTtIYSa9LekvSnZK2aqVs60i6TtIMSW9LulLS6oXtP5c0TtLXJD2X4/xF0gqldIZJeiqX9ylJh5ffVzMzMzMzM+sazX7P+s+BM4CNgZuApYD7gL2AjwO/BS6StEOO/ydgdWDnSgKSlicNW780r+8DXJjT3QQYBhwEDO9gWY8BHgA+CZwFnCVpy5xnX+BaYDrwqZznybT+/i+Ty7o9sA3wJHCTpOWqRZa0GHADsBywE7Ab8DHgqlLUDYAhedkT2BYYUUjnCODHwHGk9/6HpPfnm62U18zMzMzMzDpBI4bBt8XpEXFNKeyMwutzJO0G7A/cHRGTJY0BDgDG5jj7AjOAm/P6icDPIuLivP6spBOAc/O29rohIs7Lr0+TdCTwaWA86eLC2sD2ETEZQNIPgFtbSjAi/lFcl/RdYD9SI/wvVXbZk9QQ3z0iXs77HAg8IWmHiLi7khRwcETMyHEuBL6QXws4ATgiIq7O8Z+TtBFwKHBBPW+GmZmZmZmZtV+z96yPK65I6itpuKTHJE2TNIPUEF6rEO1SYF9JS+T1A4ArIuK93BD9JHBSHiY+I6dxIbBCeSh4Gz1aWn8FWDW/3giYWGmoZ/e3lqCkNSRdIGmCpLeAt4HlWfB4izbO+bxcCYiIJ4HXSaMIKp6tNNSrlPUjwGqkEQvF92gEsG6Ncg7LQ+vHzZ07t7XDMjPrFq6bzKxZuX4ys3o0e8/6zNL6CcBhwFHAv/P2XwFLFOJcDZwH7CnpQdJw8OPzNpEuUJxIGpZe9nYHylqe/C6YfzFEeb2t/ggsDXwPeAGYA9wN9KsRv6V8iuEtlbXy92DgoVK8D6omHDEKGAXQv3//9hynmVmnc91kZs3K9ZOZ1aPZG+tlOwBXR8QfAST1IQ37fr4SISLekXQNqUd9PeC5iLg3b/tA0iPABhHxTDeW+0lggKRVImJKDvtUSzvkUQDbA1+PiL/lsI8wvwe8mieAdSStGRGv5H02BlbK2+rxIjAV+FhEXNFaZDMzMzMzM+t8i1pj/WlgL0nbkmaSPxpYk0JjPbuU1HP+ceCy0raRwFWSXiZNvPYBsBmwRUQcT9e4kdQzfpGkHwLLkibPC2r0hEdESJoAfE3Sw6RJ406jxqPtsptI79EfJR1N+v+eC/wzIu6pp6D5gsZI4NQ8/P3vpJELA4FVIuKX9aRjZmZmZmZm7dfs96yXDSfdG34LaQK5yVSfaG0MMA3YkDwLfEVEXAd8HtiDdE/8faSZ3MsN/k4TEXNznssDD5ImaTspb57dwq5fA1YBHiEdx9nAqy3k8z7p2fXTgTtJ78NzwJfaWN6zge+SZn9/DLgD+AbwbFvSMTMzMzMzs/ZpaM96RPyFdJ91OfypGuFTgX3qSPd9Uo97re03kXqha23fpo3rq1NSJc4TwHaVdUlbk46x5nD8iHgIGFQKvqIUZ/XS+nOkR7LVSvOHVcLOI93nXwy7GLi4HNfMzMzMzMy63qI2DH6RJWk/4A1S43xd4EzggYj4d0MLZmZmZmZmZk3HjfXusxxwCunRaK+TnrH+/YaWyMzMzMzMzJqSG+vdJCIuIN2rbmZmZmZmZtaiRW2COTMzMzMzM7Mezz3rZg0yfPjwDqcxaNAgxo0b1wmlaZsRI0bMex1R9emDXapRx93ovM0WFcOHD++V9ZPrJjOrR2ecA7aV64hFk3vWzczMzMzMzJqMG+tmZmZmZmZmTcaNdTMzMzMzM7Mm48a6mZmZmZmZWZNRIyaHsp6rf//+MXPmzIbkLakh+fZ2xcmceoveNjGMpIciYlBDMu8kjaqbXC81luun7uH6qWMaee7UGf8713NttyjXTe2pYxpVR/SEusk962ZmZmZmZmZNxo11MzMzMzMzsybjxrqZmZmZmZlZk3FjvZ0kXSrpmkaXo0jS3ZLObHQ5zMzMzMzMrGPcWK9B0mhJUWXZIkc5DBjawTzqbvBL+pCkn0l6UtIsSZMk3S7py5L8fzQzMzMzM+tB+ja6AE1uDHBQKWwqQES81dKOkvpFxLudUQhJKwL3AMsAJwLjgHeBHYGRedtLnZGXmZmZmZmZNZ57ZFs2JyJeKy1zYeFe8TwE/WxJp0uaAtyRw/9/e/cdJ1dV/3/89YYQULDQlCISQEpAvrQVlG+kCAqIiArS5EtVBPzZ6EXMgoAQUYo0QZpKkaJ8iXwRQpMiBEKHdEhooUkLARICfH5/nDPJzWRndmZ3dmd29/18POYxc+85937OuZM9mTPn3DMHSpokaZakVyT9U9ICko4HvgdsXxi1H1ahHCcBKwAbRcQlEfFEREyKiAuB9YBXOjpI0p6Sxkh6K4/E/1XSsoX0wbnML+TyPSvphEL6jpIeyyP5r0m6XdLS3bukZmZmZmZm1hmPrDfWnsA5wDBgAUkbAacDewD/BhYHvpLzngSsASwK7J33vVp+QkkLAjsDf4qIaeXpEfFulfIsBBwDTACWBkYAlwGb5/SfA9sBOwHPAJ8BVs1xlwcuBw4FriWN6m/cSf3NzMzMzMysAdxZr25rSTMK23dGxDZV8k+OiMNKG5J2At4CRkbEDOBp4OGcPEPSTGBQRLxY5ZyfBj4OjKu38BHxx8LmU5J+BDwmaZkcc0VSR/6uiIhcvrtz/uVJ/z6uiojn877H6y2DmZmZmZmZ1c/T4Ku7A1i38Ph+J/nHlG3/E5gGTMnT5veQtFidZVCd+eceKLVJuk7S05LeAkbnpM/m54uANmCCpN9L2qawWN2DwO3AOElXS/qhpKUqxNkvT7cf8/7773e1uGZmDeW2ycxaldsnM6uFO+vVvRMRkwuP5zvJ/3ZxIyKmkzr5u5IWgDua1Pldpo4yvAhMB4bWcQySPgbcSBrZ3x34AvD1nDw4l+9+YAhp0bqFgL8AN0hSvjf/K8DWpBH1HwKTJH2+PFZEnBcRbRHRNmiQJ2uYWWtw22Rmrcrtk5nVwp31HhYR70fEzRFxBLAO8EnmdprfAxbs5PgPgCuBPSQtV54u6SOSFu7g0DWBJYAjIuLONiVfzAAAIABJREFUiBhPmlJffv7pEXFlROwPfBP4GrBSTouI+HdEtAMbAC+T7m83MzMzMzOzHuSv8nqQpO1J94XfAbwObAF8lLn3n08FtpC0GvAa8EZptfkyRwCbAKMl/QK4n9TR3zinfY35f7ptas7zY0nnAGuRfuatWL5D8nEPAx+QZgC8CUyTtDGwGXATqZO+AWkBurF1XwgzMzMzMzOrizvrPet10orrw0md9MnA3hFxT07/A6kT/gBptfUvA3eVnyQiXs0ryx8OHEX6AmA68ATQTrovvvyYlyTtBZwA/ITUIT8I+L9Cthn5nKuSOusPAVtHxExJb+ay/Yy0wN2zwPCIuKJrl8LMzMzMzMxq5c56BRGxVyfpu5dtz/cb6RFxB2l0utI5XgK2rLE8bwBH5kelPMPKti8n/fxakQrp5wLnVjjXE6T71c3MzMzMzKyX+Z51MzMzMzMzsxbjzrqZmZmZmZlZi3Fn3czMzMzMzKzFuLNuZmZmZmZm1mK8wJz1GxtssAFjxoxpSuy2tramxG5WXGnOOoUMHz681+M3q96l2M2os/VNEdH0f68DLXYz26eBeL3NIqLuYwbi38pA/uxkXeeRdTMzMzMzM7MW4866mZmZmZmZWYtxZ93MzMzMzMysxbizbmZmZmZmZtZivMCcWSeKC4I0Il+jNStuybHHHtvrMbfbbrumxC2P7YXmrC/rjbZjoLVPjWyb3L6Y9S31/O131jZ2ZdE+6588sm5mZmZmZmbWYtxZNzMzMzMzM2sx7qybmZmZmZmZtRh31ptMUpukkDSkAecKSTt2v1RmZmZmZmbWTP2isy5paUlnS5oqaZaklyTdIumrzS5bo0haRdIFkp7NdZwq6WpJGze7bGZmZmZmZtZY/WU1+GuAjwL7ApOBTwGbAks2s1CNIqkNuAUYBxwIjAUWBbYFfg9s0LzSmZmZmZmZWaP1+ZF1SZ8EvgwcERG3RMTTEXF/RJwSEVcU8g2WdLKk5yS9Lel+SVuVnWsNSddJelPSDEn3SFo7py0g6ZjCyPZjkrYvHDskT0PfQdIoSe9IGls+ui9pa0njJc2UdCewWif1E3Ax8BTw3xExMiKejIhHI+LXwBZVjj1J0gRJ7+aR+BGSFimkryDpfyW9lss7XtIuhfRfSno61/dFSX+qVlYzMzMzMzNrjD7fWQdm5Mc3ix3RDlxEGm3fDVgbuAQYKWkdAEnLAXcBAXwVWB84C1gwH/9T4FDg8Hz834G/SVq3LM4JwBnAOsD9wBWSFssxVgCuBUYB65JGxUd0Ur91gbWA30TEB+WJEfFGlWPfBvYBhpJG5HcBji6kn02akbB5jvEz4I1c1h2AQ/JxqwLfAO7rpKxmZmZmZmbWAH1+GnxEvC9pL+B8YD9JDwF3A1dFxGhI93sDuwJDIuKZfOiZkrYEfkjqkP6I1Ln9bkS8l/NMLIQ6BDglIi7L27+UtEnev3sh36kRMTLHPQrYg9Thvgs4AHgG+ElEBDBe0mrAr6pUcdX8PK7Wa1ISEcXzTpV0Yi7vMXnfisA1EfFI3p5SyL8i8AJwU0TMzuUeU28ZzMzMzMzMrH79YWSdiLgGWA7YDrgB2Bi4N3eWIY2SCxibp7fPkDSDdM/3KjnPesBdhY76HJI+ns9/d1nSXcCaZfseLbyelp8/lZ+HAvfmjnrJPZ1UT52kVz5Q2lHSXXkK+wzgVOCzhSynA7/I0/2Pl1S89/0qYBFgSl7Y7ruSFq4QZz9JYySNef/997taXDOzhnLbZGatyu2TmdWiX3TWASJiZkSMiojjImJj4AKgXdJgUj0D+AJplLv0GEqaJg61dYqjhn2zC2UqpZWuc1c63qXR/aH1HCTpi8AVwI2kLzHWA34BLFQo3wXASqRbBFYD/i2pPac9C6xOmnkwHfgt8ICkRctjRcR5EdEWEW2DBvX5yRpm1k+4bTKzVuX2ycxq0W866x0YS5rmvwjwEKmjvExETC57PJ/zPwgMy537eUTEdNIo+bCypGE5Tj1l2igvGlfyxU6OeTgfd6ikBcsT8wJ7Hflv4PmI+FVecG8SaWr7PCLiufwfxk7AL4H9CmkzI+L6iPg56YuOtfJ5zczMzMzMrAf1+a/yJC1JmrJ9IWkK+ltAG3AYcEvuaE+XdClwsaSDSR3zJYDNgKci4m+kxdb2B66UdALwOqmDOi4iHgZ+AxwnaRLwAOk+9S9T38+mnQscDJwm6WzSQnX7VzsgIkLS3sDNwN2Sjifdv/5RYBtgp1zfchOB5SV9jzTVfivSfftzSDqddNvARODjwNbkLx/yOgCDgNGkBfx2Js0amFRHfc3MzMzMzKwL+sPI+gzgXtJq7f8CngBOBC4jdTBL9iZN9x4BjAf+AWwCPA2QR9g3AQYDt5FG438MlG4kOoPUYR8BPA58G9ghd+Rrkhe3+w6pU/wI8HPgiBqOu4/0pcA4Uod/HHA9sCHw/yocMzKX9zTSlxhfJY2cFy1AWpF+LGmF+peAPXPaG6Tfrb+TVN8dgO9ExBTMzMzMzMysR/X5kfWImAUclR/V8s0G2vOjUp4ngK9XSPuQtGp7hyu3R8RUOrgnPSJUtn09qaNddGnFgs89bhLpC4dqecpjHQkcWZbtnEL6j6uc61rSz8yZmZmZmZlZL+sPI+tmZmZmZmZm/Yo762ZmZmZmZmYtxp11MzMzMzMzsxbjzrqZmZmZmZlZi+nzC8yZ9bSI6DRPW1sbY8aM6YXStEZcae5ahsOHD+/1+M2qd7NjmzVSLW1bdwzE9sntg9nA1Vl7097ePud1T7e/1n94ZN3MzMzMzMysxbizbmZmZmZmZtZi3Fk3MzMzMzMzazHurJuZmZmZmZm1GC8wZ2Yce+yxXT62uJhTo3kBFrOBoTttUDU92T6ZWf/TW21GT7V51Wy33XZ1xW3GAsI2P4+sm5mZmZmZmbUYd9bNzMzMzMzMWow762ZmZmZmZmYtxp11MzMzMzMzsxbTJzrrkvaSdGuzy9ERSW2SQtKQHjr/DEl7FbZD0o49ESuff6kcY7O8vbak5yUt2lMxzczMzMzMbF41d9YlrSfpA0l392SBOog7GDgeqGn5Qkmb5c7mUj1bsqZZFhjZW8Ei4jHgXuCg3oppZmZmZmY20NUzsv4D4Gzg85KG9lB5OrIjMDMi/tWLMXtV/kKiJhHxYkTM6snydOAi4ABJ/qk/MzMzMzOzXlBTZ13SR4DdgPOBq4F9O8izkaQHJc2U9JCkrxenU+c8a0q6XtJbkl6WdLmkZToJvxtwXVmstSXdIml6PtcjkjbPU9Fvy9leyfEvzsdsLelOSa9Lek3SjcUvHSQNyfl3kDRK0juSxkr6alnsrSWNz/W8E1itLH3JXK/nJL0r6QlJe5fluV3SOZJOkfQKcHfe/7mcNlPSBEnf6OA6z5kGL+nivF3+2CunS9Jhkp7MZXlM0u5l5/uCpAdK7xuwUQfvwU3AEsBmHaSZmZmZmZlZg9U6sr4j8HREPAr8GdhD0kKlREmLAf8AxgMbAIcBvymeQNKywB3A48CGwJbAYsB1kqqVYxgwpmzfZcAL+TzrAe3ATOBZYIecZy3SlPGf5u1FgdPyMZsBbwIjOxjVPgE4A1gHuB+4ItcPSSsA1wKjgHWB3wMjyo5fBHgQ+EYuw+nAHyRtUZZvd0DAl0nXcwHg76T35EvAPrleC1e8MqluyxYeRwPvMPd6HU/6YuVHwJrAr3NZts31WRS4HngKaAOOAE4pDxIR7wEPA5tWKYuZmZmZmZk1SK3Tmr9P6qQD/IvUIfwmcE3e9z1gQWDfiHgXeELSCcClhXMcADwSEYeXdkjaA3iN1FG8rzyopE8CnyB1zItWBE6JiPF5e3LhmNfyy5cj4j+l/RFxTeF48mj3dFLn/a5C0qkRMTLnOQrYg9QxvyvX4RngJxERwHhJqwG/KsR5nnm/qDhP0leAXYFbCvunRMTBhfJ8jdShXikinsn7fgbcWX5dCrHeJH3pgKRhwC+B3SLi8dwRPwj4WkSUzjFF0oakzvv1pPdtMLB3RMwAHs/v25+Z3zRgSEflkLQfsB/A4ME1z+g3M+tRbpvMrFW5fTKzWnQ6si7pc8B/k0azyZ3US0kd+JI1gMdzR71kdNmpNgA2yaubz5A0gzQSDrBKhfAfyc8zy/b/DvijpFslHS1pjRrqsYqky/KU8OnAS6T6f7Ys66OF19Py86fy81Dg3nwNSu4pi7NgLtOjkl7N9fxOB3EeKNseCjxf6qhno4EPa6jbEOBvwHER8be8e03SKP8/y675Acy93kOBR3NHvcP6FLzL3PdjHhFxXkS0RUTboEG+rd3MWoPbJjNrVW6fzKwWtbQO3yeNmj8jqbRPkKaFR8SzeTs6PnyOBUijuYd0kPZShWNezeddvLgzItolXQpsA2wFDJe0f0RcWCX+SOB54If5+X1gLGlkuWh2IU7kOpe+1BCdOwQ4mDRF/TFgBnAiczv8JW+Xbddy7vnkKfrXATdGxImFpFKZtyPNBigq1bGemEsAU7tSRjMzMzMzM6tP1ZF1pdW/9wSOJE0FLz3WIY1AlxZOGwesrbQQXcmGZad7kHQP99MRMbns8VZH8fO90mNJo8TlaZMi4oyI2Ba4gLkj/e/l5wUL9ViSNIp8YkTcHBHjgI9R+20AJWOBjVT41gL4YlmeYcDIiPhzRDwMPEnZInRVzr18vi++ZEOqvEf5PvdLgbeYd6ZD6XyzgBU7uN5PF/KsrXl/Q728PiWfJ72HZmZmZmZm1sM6mwa/LbAUcH5EPF58AFcA+xQ6jB8A5yut+L4lcFQ+R2nE/SzS/ed/VVo5fmVJW0o6T9LHqpThRlIHGEgr00s6S+n31IdI2iinj81Zns4xt5W0dB55fh34D/CDvOL6psC5pNH1epxLum/7NEmr51XZ9y/LMxHYQtKwPD3/TGClGs59M2mBvj9JWlfSl4BTOynjcNJidAcAi0taJj8+kr8AOQU4RdI+ud7rSto/3ycF6daG94ELJa2ltPL90eVB8jT75UmrwpuZmZmZmVkP66yzvi9wW0S82kHaVaSF3rbM9zxvRxo5f4i0wFp7zjcTICKmke59/xD4J/AEqQM/Kz8qOR/YWtISefsD0rT4S4AJpBXU7yEtplZa4G04aVX3l4AzI+JDYGfgv0ir0Z8FHNNJ3Pnk+8m/A2wNPAL8nLSCetHxpMXybiCtfv828y60V+ncHwLfJr0no4E/5XNVK+OmwNK5LC8UHjvn9GNI78MhpOs9irRa/pQccwZp1fpVSaPmpwBzFgAs2BW4qTAib2ZmZmZmZj2o6jTwiPhmlbSnKNzzHBH3kn5GDQBJ25NGuJ8s5JlE+hm4mkXEeEnXklYw/1WeGr9bJ8f8isIK7XnfraSp3EWLFdKn0sE93BGhsu3rSffeF11aSH+d1KGvVr7NKuyfyPw/j7ZYWZ7iNe/wPIX0IP283O+r5BkNrF+2e+7iBNLCpJH7XavFMjMzMzMzs8Zp2PKTkvYk/V73s6RO8Wmke7f/U/XA2hxGGnW23rcicEJE3N3sgpiZmZmZmQ0UjfytiE8DxwLLAi+SRp87mlJdtzz9/PRGnMvqk0f7Jza7HGZmZmZmZgNJwzrrETECGNGo85mZmZmZmZkNVI0cWTezPmr48OF15W9vb5/zOi2NYGbWdfW2QdU0s31qa2vr1Xhm1lg92WYUf/m5kW1erdra2hgzZkyvx7Xu6Ww1eDMzMzMzMzPrZe6sm5mZmZmZmbUYd9bNzMzMzMzMWow762ZmZmZmZmYtxgvMmdWguChId/L0hGpxiwstmVn/Umub06y2qZbYbqPM+q7uti2t3DZ1xO2VNYNH1s3MzMzMzMxajDvrZmZmZmZmZi3GnXUzMzMzMzOzFuPOupmZmZmZmVmLcWe9B0hql/R4E+KGpB17O66ZmZmZmZk1VqeddUkX505gSJot6SlJp0hatDcK2EedAmzahLjLAiObENfMzMzMzMwaqNafbrsZ+B9gIeDLwB+BRYEDeqhcfYKkwRHxXvn+iJgBzOjt8kTEiz0do1KdzczMzMzMrHFqnQY/KyJejIhnI+Iy4FLgWwCSFpR0gaQpkt6VNEnSYZLmnFvS2pJukTRd0luSHpG0eU5bSNIZkqZJmiXpWUknFY4dLOlkSc9JelvS/ZK2KqRvlkf9t5A0WtI7ksZIWr9YAUn7SHomp4+UdKCkKMuznaQHJM3M9TlB0uBC+tQ8xf1CSW/k6zCf8mnw1epf4fjbJZ0j6beSXpP0iqSfSlpY0lmS3sh1+Z+y4+ZMg5c0JG/vIGlUrvdYSV8tO2aTfN1mSnpJ0qlldS6V5RRJrwB3Vyq3mZmZmZmZNUZX71l/lzTKXjrH88BOwFDgaOAoYO9C/suAF4ANgfWAdmBmTvsJ8G1gF2BVYGdgQuHYi0hTyncD1gYuAUZKWqesTL8GjgDWB14FLpUkAElfIs0GOAtYF7gOOLZ4cP4C4FLgTGAtYB9gR+DEsjgHAeOBtlzPWlSrfyXfA94CNgJOAk4DrgUm5tiXAH+UtFwn5zkBOANYB7gfuELSYgCSlgduAB7K5doX2JV0LYt2B0SaVbFHJ/HMzMzMzMysm2qdBj+HpA1JHedbACJiNvDLQpapeVR7V+CCvG9F4JSIGJ+3Jxfyr0jqgN4ZEQE8A/w7x1oln2dIRDyT858paUvgh8CBhfMcExG35eOOA+4ClgeeI30hcFNEnJzzTpT0BeAHheOPBn4TERfl7SclHQ78RdKhuWwA/4qIEbVcq7I6Vqp/JU9ERHuuz+9IX0TMjojTC3U8HNgYuLrKeU6NiJH5mKNIne11SdfnQNKXCAdGxIfAOElHAH+QdExEvJPPMSUiDq4UQNJ+wH4AgwcPrpTNzKxXuW0ys1bl9snMalHryPrWkmZImgncA9wB/LiUKGn/PPX8FUkzgJ8Dny0c/zvSKPCtko6WtEYh7WJS53FinuK9bWEK/fqkEd2xOf6MfP5tgVXKyvho4fW0/Pyp/LwGcF9Z/tFl2xsAR5fFuYx0b/4yhXxjyi9ODarVv5I59clfFLwMPFbYNxt4nbl17PQ8zH9dhgL35I56yV3AYOBzhX0PVAsQEedFRFtEtA0aVPf3P2ZmPcJtk5m1KrdPZlaLWjvrd5A61KsDi0TEdyLiZQBJO5OmaF8MbJXznU3q8AGQR4jXJE3j3hh4VNI+Oe1BYAhpSvkCpOndo3KHfQEggC/k85YeQ0nT1ItmF16XRsFL9VNhXyULkKbGF+P8F2lq/iuFfG93cp75VKt/FbPLtqPCvs7ewznHFGYH1HJdivvrrrOZmZmZmZl1Xa1f5b0TEZWmbg8DRkfEmaUdefr6PCJiEjAJOEPSOcD3gQtz2lvAVcBVki4G7iWN7D5E6lAuU5ri3kXjSPeLF5VvPwisUaWe3VKt/k00FthJ0gKF0fVhwHvAk80rlpmZmZmZ2cDWiHk3E4G9JG1Duhd7F9KCcK8DSPoI6XfHrwKmAp8md/Bz+kGk+6YfJo0C7wZMB56LiHckXQpcLOlgUod6CWAz4KmI+FuNZTwDuEvSoaTR7U1Ii9oVHQf8Q9LTwJXA+8DngQ0j4rA6rsc8Oqt/k50N/Aw4W9LpwMqkxezOLNyvbmZmZmZmZr2sq6vBF/2B1Lm9jLTa+BDgt4X0D4DFSdPbJwB/J933flBOfws4lHRP+YOk6efbFDqLe5NWhB9BWoX9H6TO9tO1FjAi7iEtJvcT0j3c3wJOprAie0TcSLoXfvNclvtIi7o9U36+OnVW/6aJiOeBbUgrwT9MGum/nNpXuTczMzMzM7Me0OnIekTs1Un6e6Sf/Nq3LOm4QvpuVY4/Hzi/Svps0k+dtVdIv500Vb64b2oH+y6kMO1c0qmUrcoeETcBN1Upy5BKaWX55pS3s/pXOH6zDvZ9voN9y5Rtq/B6KmXXoDxP3r6D9PNwNZfFzMzMzMzMetaAWX4yT4EfBcwAtgT2xyPIZmZmZmZm1oIGTGcdaAMOAT4BTAGOBE5vaonMzMzMzMzMOjBgOusRsXOzy2BmZmZmZmZWiwHTWTfrjrk/Ud+xtrY2xowZ00ulaX7c9vb2Xo9pZvPqrF2C5rURzYzt9smsd9TSBlXitsmsNo1YDd7MzMzMzMzMGsiddTMzMzMzM7MW4866mZmZmZmZWYtxZ93MzMzMzMysxbizbmZmZmZmZtZi3Fk3MzMzMzMzazHurJuZmZmZmZm1GHfWzczMzMzMzFqMO+tmZmZmZmZmLcaddTMzMzMzM7MW4866mZmZmZmZWYtxZ93MzMzMzMysxSgiml0G60ckvQVMaFL4pYD/DLDYA7HOAzV2M+u8ekR8rEmxG8Jtk2P387gDObbbp+7x34pj9+fYfb5tGtSIkpgVTIiItmYEljRmoMUeiHUeqLGbXedmxG0wt02O3W/jDvTYzYjbYAOufWr2vxnHHhixm13nRpzH0+DNzMzMzMzMWow762ZmZmZmZmYtxp11a7TzHHtAxHXsgRO32bEbZaBeP8ceGHEdu28biNdvINbZsQdO3IbF9gJzZmZmZmZmZi3GI+tmZmZmZmZmLcaddaubpO9KekLSh5IqrrAoaWtJEyRNlnREYf9KkkZLmiTpr5IG1xF7CUmj8rGjJC3eQZ7NJT1ceMyU9K2cdrGkKYW0dRsVN+f7oHDu63qxzutKuie/L49K2rmQVnedK713hfSFcz0m53oNKaQdmfdPkLRVrfWsMe5BksbmOt4iacVCWofXvoGx95L0SiHG9wtpe+b3Z5KkPXsg9qmFuBMlvVFI63K9JV0o6WVJj1dIl6QzcrkelbR+Ia1bde4JA7FtqjV2ztfn26dmtU01xu537VOz2qZ8vNunBrRPbpv82ak/tk01xu4fn50iwg8/6noAQ4HVgduBtgp5FgSeBFYGBgOPAGvmtCuBXfLrc4ED6og9Ajgivz4COLmT/EsArwEfzdsXAzt2oc41xQVmVNjfo3UGVgNWza+XA14APtmVOld77wp5DgTOza93Af6aX6+Z8y8MrJTPs2AD425eeC8PKMWtdu0bGHsv4MwK/8aeys+L59eLNzJ2Wf4fAxc2qN6bAOsDj1dI/zpwAyDgi8DoRtS5px4MwLapntiV/q30dL1pUPtU499pw9umOmL3q/aplrhl+RvWNuXj3T41oH2q5W+0g38zbpv82alRsffCn5261TZ5ZN3qFhHjImJCJ9k2BCZHxFMR8R5wBbC9JAFfAa7O+S4BvlVH+O3zMbUeuyNwQ0S8U0eMRsSdozfqHBETI2JSfj0NeBlYuo4YRR2+d1XKdDWwRa7n9sAVETErIqYAk/P5GhI3Im4rvJf3Ap+ps25djl3FVsCoiHgtIl4HRgFb92DsXYHL6zh/RRFxB+lDWSXbA3+K5F7gk5KWpft17hEDtG3qSuw5+lj71Ky2qabY/bB9alrbBG6fGtg+uW3yZ6f+1jZ1JXaf/ezkzrr1lOWBZwvbz+V9SwJvRMT7Zftr9emIeAEgP3+qk/y7MP8f5wl5WsqpkhZucNxFJI2RdK/yFDJ6uc6SNiR9y/hkYXc9da703nWYJ9frTVI9azm2O3GL9iV9c1nS0bWvVa2xd8jX8WpJK3Sx3F2NTZ66thJwa2F3d+rd1bJ1t87N1N/apnpi9/X2qVltU62xi/pD+9TKbVO18rl9mpfbJn92KuoPbVNdx/f1z06DGlo06zck3Qws00HS0RHxv7WcooN9UWV/TbFriFs8z7LA2sCNhd1HAi+SGuTzgMOB4xoY97MRMU3SysCtkh4DpneQryfr/Gdgz4j4MO+uWOdKp+msvFXy1HJsd+KmjNLuQBuwaWH3fNc+Ip7s6Pguxh4JXB4RsyTtT/p2/Cv1lLsbsUt2Aa6OiA8K+7pT766Wrbt17rKB2DY1MHZfb5+a1TbVGjtl7D/tUyu3TdXKN+DaJ7dN3Yrtz059r22qNXZJn/7s5M66dSgituzmKZ4DVihsfwaYBvyHNB1kUP5WsbS/ptiSXpK0bES8kBvXl6uUYSfg7xExu3DuF/LLWZIuAg5pZNw8jYqIeErS7cB6wDW9UWdJHweuB36Rp910WucKKr13HeV5TtIg4BOkKUG1HNuduEjakvQf8aYRMau0v8K1r7Xh7TR2RLxa2DwfOLlw7GZlx95eY9yaYhfsAvyorFzdqXdXy9bdOnfZQGybGhW7H7RPzWqbao3d39qnVm6bqpVvwLVPbpu6Htufnfpk21RT7IK+/dkpunhzvR9+UH2RlEGkhRNWYu7CD2vltKuYd8GQA+uI+RvmXTBkRJW89wKbl+1bNj8LOA04qVFxSYtFLJxfLwVMYu7CMD1a53yNbwF+1kFaXXWu9t4V8vyIeRdJuTK/Xot5F0l5itoXSaklbqkxXbXWa9/A2MsWXn8buDe/XgKYksuweH69RCNj53yrA1MBNare+bghVF4kZVvmXSTlvkbUuacfDKC2qdbY1f6t9HS9aVD7VOPfacPbpjpi96v2qZa4OV+PtE352CG4fepW+1TL32ghr9umLta7xr9Tf3byZ6e669ztRsePgffIf2zPAbOAl4Ab8/7lgP8r5Ps6MDE3EEcX9q8M3EdaQOOq0h9MjbGXzA3rpPy8RN7fBvyxkG8I8DywQNnxtwKPAY8DfwEWa1RcYON87kfy8769VWdgd2A28HDhsW5X69zRe0ea/vXN/HqRXI/JuV4rF449Oh83Adimzn9bncW9Of+bK9Xxus6ufQNj/xp4Ise4DVijcOw++VpMBvZudOy83U7Zh4Xu1pt0X+IL+d/Oc6R72fYH9s/pAs7K5XqMwgfM7ta5Jx4MwLap1tjV/q30dL1pYPvU2d8KPdQ21Ri737VPncXN2+00uG3K53D71ID2CbdN/uzUD9umWmLn7Xb6+Gcn5QPNzMzMzMzMrEV4NXgzMzMzMzOzFuPOupmZmZmZmVmLcWfdzMzMzMwh6SGlAAAAQElEQVTMrMW4s25mZmZmZmbWYtxZNzMzMzMzM2sx7qybmZmZmZmZtRh31s3MzMzMzMxajDvrZmZmZmZmZi3m/wM02F+RootgiAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 1080x360 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_weights_of([0], ['Local', 'Weight Erosion', 'Global'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_weights_of([1], ['Local', 'Weight Erosion', 'Global'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_weights_of([2], ['Local', 'Weight Erosion', 'Global'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([-1.3549876,  1.2406259], dtype=float32)"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.array(model_params['Local (lr=3.0)'][2]['linear']['w'])[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([-1.230685 ,  1.1163228], dtype=float32)"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.array(model_params['Weight Erosion (lr=3.0)'][2]['linear']['w'])[1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "EaE23XWAxE2x"
   },
   "source": [
    "# Questions\n",
    "\n",
    "1. Does the net.init also initialize the output layer based on batch size?"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    " "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    " "
   ]
  }
 ],
 "metadata": {
  "colab": {
   "collapsed_sections": [],
   "name": "Jax federated learning",
   "provenance": []
  },
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}