gym_wrapper.py

# Copyright 2017 The TensorFlow Authors 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.
# ==============================================================================

"""Wrapper around gym env.

Allows for using batches of possibly identitically seeded environments.
"""

import gym
import numpy as np
import random

from six.moves import xrange
import env_spec


def get_env(env_str):
  return gym.make(env_str)


class GymWrapper(object):

  def __init__(self, env_str, distinct=1, count=1, seeds=None):
    self.distinct = distinct
    self.count = count
    self.total = self.distinct * self.count
    self.seeds = seeds or [random.randint(0, 1e12)
                           for _ in xrange(self.distinct)]

    self.envs = []
    for seed in self.seeds:
      for _ in xrange(self.count):
        env = get_env(env_str)
        env.seed(seed)
        if hasattr(env, 'last'):
          env.last = 100  # for algorithmic envs
        self.envs.append(env)

    self.dones = [True] * self.total
    self.num_episodes_played = 0

    one_env = self.get_one()
    self.use_action_list = hasattr(one_env.action_space, 'spaces')
    self.env_spec = env_spec.EnvSpec(self.get_one())

  def get_seeds(self):
    return self.seeds

  def reset(self):
    self.dones = [False] * self.total
    self.num_episodes_played += len(self.envs)

    # reset seeds to be synchronized
    self.seeds = [random.randint(0, 1e12) for _ in xrange(self.distinct)]
    counter = 0
    for seed in self.seeds:
      for _ in xrange(self.count):
        self.envs[counter].seed(seed)
        counter += 1

    return [self.env_spec.convert_obs_to_list(env.reset())
            for env in self.envs]

  def reset_if(self, predicate=None):
    if predicate is None:
      predicate = self.dones
    if self.count != 1:
      assert np.all(predicate)
      return self.reset()
    self.num_episodes_played += sum(predicate)
    output = [self.env_spec.convert_obs_to_list(env.reset())
              if pred else None
              for env, pred in zip(self.envs, predicate)]
    for i, pred in enumerate(predicate):
      if pred:
        self.dones[i] = False
    return output

  def all_done(self):
    return all(self.dones)

  def step(self, actions):

    def env_step(env, action):
      action = self.env_spec.convert_action_to_gym(action)
      obs, reward, done, tt = env.step(action)
      obs = self.env_spec.convert_obs_to_list(obs)
      return obs, reward, done, tt

    actions = zip(*actions)
    outputs = [env_step(env, action)
               if not done else (self.env_spec.initial_obs(None), 0, True, None)
               for action, env, done in zip(actions, self.envs, self.dones)]
    for i, (_, _, done, _) in enumerate(outputs):
      self.dones[i] = self.dones[i] or done
    obs, reward, done, tt = zip(*outputs)
    obs = [list(oo) for oo in zip(*obs)]
    return [obs, reward, done, tt]

  def get_one(self):
    return random.choice(self.envs)

  def __len__(self):
    return len(self.envs)