forked from google-research/google-research
-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_env.py
144 lines (125 loc) · 4.94 KB
/
run_env.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
# coding=utf-8
# Copyright 2019 The Google Research Authors.
#
# 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.
"""Library function for stepping/evaluating a policy in a Gym environment.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import datetime
import os
import gin
import numpy as np
import PIL.Image as Image
import six
import tensorflow as tf
def encode_image_array_as_png_str(image):
"""Encodes a numpy array into a PNG string.
Args:
image: a numpy array with shape [height, width, 3].
Returns:
PNG encoded image string.
"""
image_pil = Image.fromarray(np.uint8(image))
output = six.BytesIO()
image_pil.save(output, format='PNG')
png_string = output.getvalue()
output.close()
return png_string
@gin.configurable(blacklist=['task', 'num_episodes', 'global_step', 'tag'])
def run_env(env,
policy=None,
explore_schedule=None,
episode_to_transitions_fn=None,
replay_writer=None,
root_dir=None,
task=0,
global_step=0,
num_episodes=100,
tag='collect'):
"""Runs agent+env loop num_episodes times and log performance + collect data.
Interpolates between an exploration policy and greedy policy according to a
explore_schedule. Run this function separately for collect/eval.
Args:
env: Gym environment.
policy: Policy to collect/evaluate.
explore_schedule: Exploration schedule that defines a `value(t)` function
to compute the probability of exploration as a function of global step t.
episode_to_transitions_fn: Function that converts episode data to transition
protobufs (e.g. TFExamples).
replay_writer: Instance of a replay writer that writes a list of transition
protos to disk (optional).
root_dir: Root directory of the experiment summaries and data collect. If
replay_writer is specified, data is written to the `policy_*` subdirs.
Setting root_dir=None results in neither summaries or transitions being
saved to disk.
task: Task number for replica trials for a given experiment.
global_step: Training step corresponding to policy checkpoint.
num_episodes: Number of episodes to run.
tag: String prefix for evaluation summaries and collect data.
"""
episode_rewards = []
episode_q_values = collections.defaultdict(list)
if root_dir and replay_writer:
timestamp = datetime.datetime.now().strftime('%Y-%m-%d-%H-%M-%S')
record_prefix = os.path.join(root_dir, 'policy_%s' % tag,
'gs%d_t%d_%s' % (global_step, task, timestamp))
if root_dir:
summary_dir = os.path.join(root_dir, 'live_eval_%d' % task)
summary_writer = tf.summary.FileWriter(summary_dir)
if replay_writer:
replay_writer.open(record_prefix)
for ep in range(num_episodes):
done, env_step, episode_reward, episode_data = (False, 0, 0.0, [])
policy.reset()
obs = env.reset()
if explore_schedule:
explore_prob = explore_schedule.value(global_step)
else:
explore_prob = 0
while not done:
action, policy_debug = policy.sample_action(obs, explore_prob)
if policy_debug and 'q' in policy_debug:
episode_q_values[env_step].append(policy_debug['q'])
new_obs, rew, done, env_debug = env.step(action)
env_step += 1
episode_reward += rew
episode_data.append((obs, action, rew, new_obs, done, env_debug))
obs = new_obs
if done:
tf.logging.info('Episode %d reward: %f' % (ep, episode_reward))
episode_rewards.append(episode_reward)
if replay_writer:
transitions = episode_to_transitions_fn(episode_data)
replay_writer.write(transitions)
if episode_rewards and len(episode_rewards) % 10 == 0:
tf.logging.info('Average %d collect episodes reward: %f' %
(len(episode_rewards), np.mean(episode_rewards)))
tf.logging.info('Closing environment.')
env.close()
if replay_writer:
replay_writer.close()
if root_dir:
summary_values = [
tf.Summary.Value(
tag='%s/episode_reward' % tag,
simple_value=np.mean(episode_rewards))
]
for step, q_values in episode_q_values.items():
summary_values.append(
tf.Summary.Value(
tag='%s/Q/%d' % (tag, step), simple_value=np.mean(q_values)))
summary = tf.Summary(value=summary_values)
summary_writer.add_summary(summary, global_step)