TF_Dev_Summit_2019_S4TF_Prezo_02_Interoperability.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "TF Dev Summit - 2019 - S4TF Prezo - 02 Interoperability",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "swift",
      "display_name": "Swift"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-EmSwFktm8eO",
        "colab_type": "text"
      },
      "source": [
        "# Seamless Python Interoperability\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kZRlD4utdPuX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import TensorFlow\n",
        "import Python\n",
        "%include \"EnableIPythonDisplay.swift\""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2_mVGpk3onxv",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 286
        },
        "outputId": "d3ccca40-1f44-4759-8e97-c4b4c85a5069"
      },
      "source": [
        "let plt = Python.import(\"matplotlib.pyplot\")\n",
        "let np = Python.import(\"numpy\")\n",
        "\n",
        "IPythonDisplay.shell.enable_matplotlib(\"inline\")\n",
        "\n",
        "let x = np.linspace(0, 10, 100)\n",
        "\n",
        "plt.plot(x, np.sin(x))\n",
        "plt.plot(x, np.cos(x))\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAD8CAYAAABzTgP2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nGGNMDZYYjDHG1GCJwRhjTA3/BXzX/LCh06jdAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "None\n"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 2
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PydQfPjQRuUE",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hPPeGNApt4FE",
        "colab_type": "text"
      },
      "source": [
        "# Demo: RL with OpenAI Gym\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hJ4J8BiCIpI7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "let gym = Python.import(\"gym\")\n",
        "\n",
        "/// Model parameters and hyper parameters.\n",
        "let hiddenSize = 128\n",
        "let batchSize = 16\n",
        "/// Controls the amount of good/long episodes to retain for training.\n",
        "let percentile = 70"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ogCNnuIfuZhM",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "/// A simple two layer dense net.\n",
        "struct Net: Layer {\n",
        "    var l1, l2: Dense<Float>\n",
        "\n",
        "    init(observationSize: Int, hiddenSize: Int, actionCount: Int) {\n",
        "        l1 = Dense<Float>(inputSize: observationSize, outputSize: hiddenSize, activation: relu)\n",
        "        l2 = Dense<Float>(inputSize: hiddenSize, outputSize: actionCount)\n",
        "    }\n",
        "\n",
        "    @differentiable\n",
        "    func callAsFunction(_ input: Tensor<Float>) -> Tensor<Float> {\n",
        "        return input.sequenced(through: l1, l2)\n",
        "    }\n",
        "}"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "w9x8OuioucPl",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "/// An episode is a list of steps, where each step records the observation from\n",
        "/// env and the action taken. They will serve respectively as the input and\n",
        "/// target (label) of the neural net training.\n",
        "struct Episode {\n",
        "    struct Step {\n",
        "        let observation: Tensor<Float>\n",
        "        let action: Int32\n",
        "    }\n",
        "\n",
        "    let steps: [Step]\n",
        "    let reward: Float\n",
        "}\n",
        "\n",
        "/// Filters out bad/short episodes before we feed them as neural net training data.\n",
        "func filteringBatch(\n",
        "    episodes: [Episode],\n",
        "    actionCount: Int\n",
        ") -> (input: Tensor<Float>,\n",
        "      target: Tensor<Float>,\n",
        "      episodeCount: Int,\n",
        "      meanReward: Float) {\n",
        "    let rewards = episodes.map { $0.reward }\n",
        "    let rewardBound = Float(np.percentile(rewards, percentile))!\n",
        "    print(\"rewardBound = \\(rewardBound)\")\n",
        "\n",
        "    var input = Tensor<Float>(0.0)\n",
        "    var target = Tensor<Float>(0.0)\n",
        "    var totalReward: Float = 0.0\n",
        "\n",
        "    var retainedEpisodeCount = 0\n",
        "    for episode in episodes {\n",
        "        if episode.reward < rewardBound {\n",
        "            continue\n",
        "        }\n",
        "\n",
        "        let observationTensor = Tensor<Float>(episode.steps.map { $0.observation })\n",
        "        let actionTensor = Tensor<Int32>(episode.steps.map { $0.action })\n",
        "        let oneHotLabels = Tensor<Float>(oneHotAtIndices: actionTensor, depth: actionCount)\n",
        "\n",
        "        if retainedEpisodeCount == 0 {\n",
        "            input = observationTensor\n",
        "            target = oneHotLabels\n",
        "        } else {\n",
        "            input = input.concatenated(with: observationTensor)\n",
        "            target = target.concatenated(with: oneHotLabels)\n",
        "        }\n",
        "\n",
        "        totalReward += episode.reward\n",
        "        retainedEpisodeCount += 1\n",
        "    }\n",
        "  \n",
        "    return (input, target, retainedEpisodeCount, totalReward / Float(retainedEpisodeCount))\n",
        "}"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Mg9ig-F7uecY",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "func nextBatch(env: PythonObject,\n",
        "               net: Net,\n",
        "               batchSize: Int,\n",
        "               actionCount: Int\n",
        ") -> [Episode] {\n",
        "    var observationNumpy = env.reset()\n",
        "\n",
        "    var episodes = [Episode]()\n",
        "\n",
        "    withLearningPhase(.inference) {  \n",
        "        // Build up a batch of observations and actions.\n",
        "        for _ in 0..<batchSize {\n",
        "            var steps = [Episode.Step]()\n",
        "            var episodeReward: Float = 0.0\n",
        "\n",
        "            while true {\n",
        "                let observationPython = Tensor<Double>(numpy: observationNumpy)!\n",
        "                let actionProbabilities =\n",
        "                    softmax(net(Tensor(observationPython).reshaped(to: [1, 4])))\n",
        "                let actionProbabilitiesPython = actionProbabilities[0].makeNumpyArray()\n",
        "                let len = Python.len(actionProbabilitiesPython)\n",
        "                assert(actionCount == Int(len))\n",
        "      \n",
        "                let actionPython = np.random.choice(len, p: actionProbabilitiesPython)\n",
        "                let (nextObservation, reward, isDone, _) = env.step(actionPython).tuple4\n",
        "\n",
        "                steps.append(Episode.Step(observation: Tensor<Float>(observationPython),\n",
        "                                          action: Int32(actionPython)!))\n",
        "\n",
        "                episodeReward += Float(reward)!\n",
        "\n",
        "                if isDone == true {\n",
        "                    episodes.append(Episode(steps: steps, reward: episodeReward))\n",
        "                    observationNumpy = env.reset()\n",
        "                    break\n",
        "                } else {\n",
        "                    observationNumpy = nextObservation\n",
        "                }\n",
        "            }\n",
        "        }\n",
        "    }\n",
        "    return episodes\n",
        "\n",
        "}"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OzAYq5lbuoJB",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "let env = gym.make(\"CartPole-v0\")\n",
        "let observationSize = Int(env.observation_space.shape[0])!\n",
        "let actionCount = Int(env.action_space.n)!\n",
        "\n",
        "var meanRewards = [Float]()\n",
        "\n",
        "var net = Net(\n",
        "    observationSize: observationSize,\n",
        "    hiddenSize: hiddenSize,\n",
        "    actionCount: actionCount)\n",
        "let optimizer = Adam(for: net, learningRate: 0.01)\n",
        "Context.local.learningPhase = .training\n",
        "\n",
        "for batchIndex in 0... {\n",
        "    print(\"Processing mini batch \\(batchIndex)\")\n",
        "  \n",
        "    let episodes = nextBatch(\n",
        "        env: env, net: net, batchSize: batchSize, actionCount: actionCount)\n",
        "    let (input, target, episodeCount, meanReward) = \n",
        "        filteringBatch(episodes: episodes, actionCount: actionCount)\n",
        "\n",
        "    let gradients = gradient(at: net) { model -> Tensor<Float> in\n",
        "        let logits = model(input)\n",
        "        let loss = softmaxCrossEntropy(logits: logits, oneHotLabels: target)\n",
        "        print(\"loss is \\(loss)\")\n",
        "        return loss\n",
        "    }\n",
        "    optimizer.update(&net, along: gradients)\n",
        "\n",
        "    print(\"It has episode count \\(episodeCount) and mean reward \\(meanReward)\")\n",
        "    meanRewards.append(meanReward)\n",
        "\n",
        "    if meanReward > 199 {\n",
        "        print(\"Solved\")\n",
        "        break\n",
        "    }\n",
        "}"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BBYAOCxHuqZb",
        "colab_type": "code",
        "outputId": "c85948aa-eb30-4425-d039-308586677737",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 286
        }
      },
      "source": [
        "plt.plot(meanRewards)\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAD8CAYAAAB5Pm/hAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "None\n"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 30
        }
      ]
    }
  ]
}