run.py

# python3
# pylint: disable=g-bad-file-header
# Copyright 2019 DeepMind Technologies Limited. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or  implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
"""Run a Dqn agent instance (using JAX) on a bsuite experiment."""

from absl import app
from absl import flags

import bsuite
from bsuite import sweep

from bsuite.baselines import experiment
from bsuite.baselines.jax import boot_dqn
from bsuite.baselines.utils import pool

import haiku as hk
from jax import lax
import jax.numpy as jnp
import optax

# Internal imports.

flags.DEFINE_integer('num_ensemble', 20, 'Size of ensemble.')

# Experiment flags.
flags.DEFINE_string(
    'bsuite_id', 'catch/0', 'BSuite identifier. '
    'This global flag can be used to control which environment is loaded.')
flags.DEFINE_string('save_path', '/tmp/bsuite', 'where to save bsuite results')
flags.DEFINE_enum('logging_mode', 'csv', ['csv', 'sqlite', 'terminal'],
                  'which form of logging to use for bsuite results')
flags.DEFINE_boolean('overwrite', False, 'overwrite csv logging if found')
flags.DEFINE_integer('num_episodes', None, 'Number of episodes to run for.')
flags.DEFINE_boolean('verbose', True, 'whether to log to std output')

FLAGS = flags.FLAGS


def run(bsuite_id: str) -> str:
  """Runs a DQN agent on a given bsuite environment, logging to CSV."""

  env = bsuite.load_and_record(
      bsuite_id=bsuite_id,
      save_path=FLAGS.save_path,
      logging_mode=FLAGS.logging_mode,
      overwrite=FLAGS.overwrite,
  )
  action_spec = env.action_spec()

  # Define network.
  prior_scale = 5.
  hidden_sizes = [50, 50]
  def network(inputs: jnp.ndarray) -> jnp.ndarray:
    """Simple Q-network with randomized prior function."""
    net = hk.nets.MLP([*hidden_sizes, action_spec.num_values])
    prior_net = hk.nets.MLP([*hidden_sizes, action_spec.num_values])
    x = hk.Flatten()(inputs)
    return net(x) + prior_scale * lax.stop_gradient(prior_net(x))

  optimizer = optax.adam(learning_rate=1e-3)

  agent = boot_dqn.BootstrappedDqn(
      obs_spec=env.observation_spec(),
      action_spec=action_spec,
      network=network,
      optimizer=optimizer,
      num_ensemble=FLAGS.num_ensemble,
      batch_size=128,
      discount=.99,
      replay_capacity=10000,
      min_replay_size=128,
      sgd_period=1,
      target_update_period=4,
      mask_prob=1.0,
      noise_scale=0.,
  )

  num_episodes = FLAGS.num_episodes or getattr(env, 'bsuite_num_episodes')
  experiment.run(
      agent=agent,
      environment=env,
      num_episodes=num_episodes,
      verbose=FLAGS.verbose)

  return bsuite_id


def main(_):
  # Parses whether to run a single bsuite_id, or multiprocess sweep.
  bsuite_id = FLAGS.bsuite_id

  if bsuite_id in sweep.SWEEP:
    print(f'Running single experiment: bsuite_id={bsuite_id}.')
    run(bsuite_id)

  elif hasattr(sweep, bsuite_id):
    bsuite_sweep = getattr(sweep, bsuite_id)
    print(f'Running sweep over bsuite_id in sweep.{bsuite_sweep}')
    FLAGS.verbose = False
    pool.map_mpi(run, bsuite_sweep)

  else:
    raise ValueError(f'Invalid flag: bsuite_id={bsuite_id}.')


if __name__ == '__main__':
  app.run(main)