-
Notifications
You must be signed in to change notification settings - Fork 11
/
Copy pathppo.py
184 lines (156 loc) · 7.26 KB
/
ppo.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
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
# Copyright 2022 Twitter, Inc.
# SPDX-License-Identifier: Apache-2.0
import hydra.utils
import torch
import torch as th
from torch import nn
from utils import to_numpy
from logging_metrics import AverageMetrics
class PPO():
def __init__(self,
device,
actor_critic_model,
optimizer_fn=th.optim.Adam,
optimizer_kwargs={'lr': 5e-4},
clipping=0.2,
value_clipping=None,
ent_coeff=0.0,
val_coeff=0.5,
max_gradient_norm=None,
predictions_logging=False,
optimizer_scheduler=None,):
self.device = device
self.ac_model = actor_critic_model.to(device=device)
self.optimizer = optimizer_fn(self.ac_model.parameters(),
**optimizer_kwargs)
if optimizer_scheduler is not None:
self.scheduler = hydra.utils.instantiate(optimizer_scheduler, optimizer=self.optimizer)
else:
self.scheduler = None
self.clipping = clipping
self.value_clipping = value_clipping
self.ent_coeff = ent_coeff
self.val_coeff = val_coeff
self.max_gradient_norm = max_gradient_norm
self.predictions_logging = predictions_logging
self.metrics = AverageMetrics('actor_loss',
'entropy_loss',
'value_loss', )
if predictions_logging:
self.metrics.add('mean_logprobs', 'predicted_values', 'discounted_returns')
self.debug_metrics = AverageMetrics('average_returns',
'std_returns',
'std_advantages')
def save(self, dir, step, preprocessor=None):
checkpoint = '{}/checkpoint-{}.pt'.format(dir, step)
agent_params = self.ac_model.state_dict()
optimizer_params = self.optimizer.state_dict()
if preprocessor is not None:
th.save({'agent': agent_params,
'optimizer': optimizer_params,
'preprocessor': preprocessor}, checkpoint)
else:
th.save({'agent': agent_params,
'optimizer': optimizer_params, }, checkpoint)
return checkpoint
def load(self, path):
params = th.load(path)
self.ac_model.load_state_dict(params['agent'])
self.optimizer.load_state_dict(params['optimizer'])
if 'preprocessor' in params:
return params['preprocessor']
def get_action_logprob_value(self, obs):
with th.no_grad():
act, logprob, value = self.ac_model.get_action_logprob_value(obs)
return act, logprob, value
def get_value(self, obs):
with th.no_grad():
return self.ac_model.get_value(obs)
def act(self, obs, det=False):
with th.no_grad():
return self.ac_model.get_action(obs, det=det).cpu().numpy()
def get_losses(self, obs, act, old_logprobs, old_values, returns,
advantages):
metrics_dict = dict()
logprobs, entropies, values = self.ac_model.get_logprob_entropy_value(
obs=obs, act=act)
with th.no_grad():
adv_std, adv_mean = th.std_mean(advantages)
norm_advantages = (advantages - adv_mean) / (adv_std + 1e-7)
ratio = th.exp(logprobs - old_logprobs)
clipped_ratio = th.clamp(ratio, min=1 - self.clipping,
max=1 + self.clipping)
pessimistic_adv = th.minimum(ratio * norm_advantages,
clipped_ratio * norm_advantages)
actor_loss = -1 * pessimistic_adv.mean()
entropy_loss = -1 * entropies.mean()
value_losses = (values - returns).pow(2)
if self.value_clipping is not None:
clipped_values = old_values + th.clamp(values - old_values,
min=-self.value_clipping,
max=self.value_clipping)
clipped_value_losses = (clipped_values - returns).pow(2)
value_losses = th.maximum(value_losses, clipped_value_losses)
value_loss = 0.5 * value_losses.mean()
if self.predictions_logging:
metrics_dict['mean_logprobs'] = logprobs.mean()
metrics_dict['predicted_values'] = values.mean()
metrics_dict['discounted_returns'] = returns.mean()
return actor_loss, entropy_loss, value_loss, metrics_dict
def learn(self, buffer, epochs, batch_size, current_steps=None):
if current_steps is not None:
self.ac_model.update(current_steps=current_steps)
for e in range(epochs):
for batch in buffer.get_batches(batch_size):
actor_loss, entropy_loss, value_loss, metrics_dict = self.get_losses(*batch)
total_loss = (actor_loss + self.ent_coeff * entropy_loss +
self.val_coeff * value_loss)
self.ac_model.zero_grad(set_to_none=True)
total_loss.backward()
if self.max_gradient_norm is not None:
nn.utils.clip_grad_norm_(self.ac_model.parameters(),
self.max_gradient_norm)
self.optimizer.step()
if self.scheduler is not None:
self.scheduler.step()
metrics_dict = {k: to_numpy(t) for k, t in metrics_dict.items()}
self.metrics.update(actor_loss=to_numpy(actor_loss),
entropy_loss=to_numpy(entropy_loss),
value_loss=to_numpy(value_loss),
**metrics_dict)
def return_and_reset_metrics(self, ):
metrics = self.metrics.get()
self.metrics.reset()
return metrics
class ModularPPO(PPO):
def __init__(self,
device,
actor_critic_model,
optimizer,
clipping=0.2,
value_clipping=None,
ent_coeff=0.0,
val_coeff=0.5,
max_gradient_norm=None,
predictions_logging=False,
optimizer_scheduler=None,):
self.metrics = AverageMetrics('actor_loss',
'entropy_loss',
'value_loss',)
if predictions_logging:
self.metrics.add('mean_logprobs',
'predicted_values',
'discounted_returns',)
self.device = device
self.ac_model = hydra.utils.instantiate(actor_critic_model, _recursive_=False).to(device=device)
self.optimizer = hydra.utils.instantiate(optimizer, params=self.ac_model.parameters())
if optimizer_scheduler is not None:
self.scheduler = hydra.utils.instantiate(optimizer_scheduler, optimizer=self.optimizer)
else:
self.scheduler = None
self.clipping = clipping
self.value_clipping = value_clipping
self.ent_coeff = ent_coeff
self.val_coeff = val_coeff
self.max_gradient_norm = max_gradient_norm
self.predictions_logging = predictions_logging