-
Notifications
You must be signed in to change notification settings - Fork 128
/
ac.py
481 lines (397 loc) · 17.5 KB
/
ac.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
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
'''
Actor-Critic, actually Advantage Actor-Critic (A2C).
Policy loss in Vanilla Actor-Critic is: -log_prob(a)*Q(s,a) ,
Policy loss in A2C is: -log_prob(a)*[Q(s,a)-V(s)], while Adv(s,a)=Q(s,a)-V(s)=r+gamma*V(s')-V(s)=TD_error ,
and in this implementation we provide another approach that the V(s') is replaced by R(s'),
which is derived from the rewards in the episode for on-policy update without evaluation.
Discrete and Non-deterministic
'''
import math
import random
import gym
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torch.distributions import Normal
from torch.distributions import Categorical
from collections import namedtuple
from IPython.display import clear_output
import matplotlib.pyplot as plt
from matplotlib import animation
from IPython.display import display
from reacher import Reacher
# use_cuda = torch.cuda.is_available()
# device = torch.device("cuda" if use_cuda else "cpu")
# print(device)
GPU = True
device_idx = 0
if GPU:
device = torch.device("cuda:" + str(device_idx) if torch.cuda.is_available() else "cpu")
else:
device = torch.device("cpu")
print(device)
DISCRETE = True # discrete actions if ture, else continuous
DETERMINISTIC = False # deterministic actions if true, like DDPG or DQN's argmax, else non-deterministic (sampling)
if DISCRETE:
# each output node corresponds to one possible action,
# the output dim = possible action values (only one action)
pass
else:
# the output dim = dim of action
pass
if DETERMINISTIC:
# no need of sampling, directly output actions
pass
else:
# output the mean and (log-)variance for Gaussian prior, then sampling
pass
class ReplayBuffer:
def __init__(self, capacity):
self.capacity = capacity
self.buffer = []
self.position = 0
def push(self, state, action, reward, next_state, done):
if len(self.buffer) < self.capacity:
self.buffer.append(None)
self.buffer[self.position] = (state, action, reward, next_state, done)
self.position = int((self.position + 1) % self.capacity) # as a ring buffer
def sample(self, batch_size):
batch = random.sample(self.buffer, batch_size)
state, action, reward, next_state, done = map(np.stack, zip(*batch)) # stack for each element
'''
the * serves as unpack: sum(a,b) <=> batch=(a,b), sum(*batch) ;
zip: a=[1,2], b=[2,3], zip(a,b) => [(1, 2), (2, 3)] ;
the map serves as mapping the function on each list element: map(square, [2,3]) => [4,9] ;
np.stack((1,2)) => array([1, 2])
'''
return state, action, reward, next_state, done
def __len__(self):
return len(self.buffer)
class ActorNetwork(nn.Module):
def __init__(self, input_dim, output_dim, hidden_dim, init_w=3e-3):
super(ActorNetwork, self).__init__()
self.saved_logprobs = [] # this is critical! have to save the values inside the model to keep track of its gradients
self.saved_entropies = []
self.linear1 = nn.Linear(input_dim, hidden_dim)
self.linear2 = nn.Linear(hidden_dim, hidden_dim)
if DISCRETE: # e.g. DQN for deterministic and Actor-Critic for non-deterministic
self.linear3 = nn.Linear(hidden_dim, output_dim) # output dim = possible action values
# weights initialization
self.linear3.weight.data.uniform_(-init_w, init_w)
self.linear3.bias.data.uniform_(-init_w, init_w)
elif not DISCRETE and DETERMINISTIC: # e.g. DDPG
self.linear3 = nn.Linear(hidden_dim, output_dim) # output dim = dim of action
# weights initialization
self.linear3.weight.data.uniform_(-init_w, init_w)
self.linear3.bias.data.uniform_(-init_w, init_w)
elif not DISCRETE and not DETERMINISTIC: # e.g. REINFORCE, Actor-Critic, PPO for continuous case
self.mean_linear = nn.Linear(hidden_dim, output_dim) # output dim = dim of action
self.log_std_linear = nn.Linear(hidden_dim, output_dim)
# weights initialization
self.mean_linear.weight.data.uniform_(-init_w, init_w)
self.mean_linear.bias.data.uniform_(-init_w, init_w)
self.log_std_linear.weight.data.uniform_(-init_w, init_w)
self.log_std_linear.bias.data.uniform_(-init_w, init_w)
def forward(self, state):
x = F.relu(self.linear1(state))
x = F.relu(self.linear2(x))
if DISCRETE and DETERMINISTIC:
x = torch.max(self.linear3(x), dim=-1)
return x
elif DISCRETE and not DETERMINISTIC:
x = F.softmax(self.linear3(x), dim=-1)
return x
elif not DISCRETE and not DETERMINISTIC:
self.log_std_min=-20
self.log_std_max=2
mean = self.mean_linear(x)
log_std = self.log_std_linear(x)
log_std = torch.clamp(log_std, self.log_std_min, self.log_std_max)
return mean, log_std
else:
x = self.linear3(x)
return x
def select_action(self, state):
'''
only select action without the purpose of gradients flow, for interaction with env to
generate samples
'''
if DETERMINISTIC:
action = self.forward(state)
if DISCRETE and not DETERMINISTIC:
probs = self.forward(state)
m = Categorical(probs)
action = m.sample()
if not DISCRETE and not DETERMINISTIC:
self.action_range = 30.
mean, log_std = self.forward(state)
std = log_std.exp()
normal = Normal(0, 1)
z = normal.sample().to(device)
action = self.action_range* torch.tanh(mean + std*z)
return action.detach()
def evaluate_action(self, state):
'''
evaluate action within GPU graph, for gradients flowing through it
'''
state = torch.FloatTensor(state).unsqueeze(0).to(device) # state dim: (N, dim of state)
if DETERMINISTIC:
action = self.forward(state)
return action.detach().cpu().numpy()
elif DISCRETE and not DETERMINISTIC: # actor-critic (discrete)
probs = self.forward(state)
m = Categorical(probs)
action = m.sample().to(device)
log_prob = m.log_prob(action)
return action.detach().cpu().numpy(), log_prob.squeeze(0), m.entropy().mean()
elif not DISCRETE and not DETERMINISTIC: # soft actor-critic (continuous)
self.action_range = 30.
self.epsilon = 1e-6
mean, log_std = self.forward(state)
std = log_std.exp()
normal = Normal(0, 1)
z = normal.sample().to(device)
action0 = torch.tanh(mean + std*z.to(device)) # TanhNormal distribution as actions; reparameterization trick
action = self.action_range * action0
log_prob = Normal(mean, std).log_prob(mean+ std*z.to(device)) - torch.log(1. - action0.pow(2) + self.epsilon) - np.log(self.action_range)
log_prob = log_prob.sum(dim=1, keepdim=True)
# print('mean: ', mean, 'log_std: ', log_std)
# return action.item(), log_prob, z, mean, log_std
return action.detach().cpu().numpy().squeeze(0), log_prob.squeeze(0), Normal(mean, std).entropy().mean()
class CriticNetwork(nn.Module):
def __init__(self, input_dim, hidden_dim, init_w=3e-3):
super(CriticNetwork, self).__init__()
self.saved_values = [] # this is critical! have to save the values inside the model to keep track of its gradients
self.saved_nextvalues = []
self.linear1 = nn.Linear(input_dim, hidden_dim)
self.linear2 = nn.Linear(hidden_dim, hidden_dim)
self.linear3 = nn.Linear(hidden_dim, 1)
# weights initialization
self.linear3.weight.data.uniform_(-init_w, init_w)
self.linear3.bias.data.uniform_(-init_w, init_w)
def forward(self, state):
state = torch.FloatTensor(state).to(device)
x = F.relu(self.linear1(state))
x = F.relu(self.linear2(x))
x = self.linear3(x)
return x
# class QNetwork(nn.Module):
# def __init__(self, input_dim, hidden_dim, init_w=3e-3):
# super(QNetwork, self).__init__()
# self.linear1 = nn.Linear(input_dim, hidden_dim)
# self.linear2 = nn.Linear(hidden_dim, hidden_dim)
# self.linear3 = nn.Linear(hidden_dim, 1)
# self.linear3.weight.data.uniform_(-init_w, init_w)
# self.linear3.bias.data.uniform_(-init_w, init_w)
# def forward(self, state, action):
# x = torch.cat([state, action], 1) # the dim 0 is number of samples
# x = F.relu(self.linear1(x))
# x = F.relu(self.linear2(x))
# x = self.linear3(x)
# return x
def Update0(rewards, gamma=0.99, entropy_lambda=1e-3):
''' update with R(s') instead of V(s') in the TD-error;
with entropy boosting exploration
'''
# print('sets: ', actions)
# print('rewards: ', rewards)
R = 0
policy_losses = []
value_losses = []
rewards_ = []
eps = np.finfo(np.float32).eps.item()
for r in rewards[::-1]:
R = r + gamma * R
rewards_.insert(0, R)
rewards_ = torch.tensor(rewards_).to(device)
rewards_ = (rewards_ - rewards_.mean()) / (rewards_.std() + eps)
# print('rewards: ', rewards)
# print('rewards_: ', rewards_)
for log_prob, value, r in zip(actor_net.saved_logprobs, critic_net.saved_values, rewards_):
value_losses.append(F.smooth_l1_loss(value, torch.tensor([r]).to(device)))
td_error = r - value.detach().item() # value gradients flow only through the critic
policy_losses.append(-log_prob * td_error)
# print('policy losses: ', policy_losses)
# print('value losses: ', value_losses)
actor_optimizer.zero_grad()
policy_loss=torch.stack(policy_losses).sum() - entropy_lambda * torch.stack(actor_net.saved_entropies).sum()
policy_loss.backward()
actor_optimizer.step()
critic_optimizer.zero_grad()
value_loss=torch.stack(value_losses).sum()
value_loss.backward()
# print('loss: ', policy_loss, value_loss)
critic_optimizer.step()
del actor_net.saved_logprobs[:]
del critic_net.saved_values[:]
del actor_net.saved_entropies[:]
def Update1(rewards, gamma=0.99):
''' update with V(s') in the TD-error'''
policy_losses = []
value_losses = []
value_criterion = nn.MSELoss()
rewards = torch.tensor(rewards).to(device)
for log_prob, state_value, next_state_value, r in zip(actor_net.saved_logprobs, critic_net.saved_values, critic_net.saved_nextvalues, rewards):
# value_losses.append(F.smooth_l1_loss(state_value, r + gamma * next_state_value.detach_())) # detach the next_state_value, only BP through state_value
value_losses.append(value_criterion(state_value, r + gamma * next_state_value.detach_()))
state_value.detach_() # detach in place
policy_losses.append(-log_prob * (r + gamma * next_state_value - state_value)) # only BP through the log_prob for actor update
# print('policy losses: ', policy_losses)
# print('value losses: ', value_losses)
actor_optimizer.zero_grad()
policy_loss=torch.stack(policy_losses).sum()
policy_loss.backward()
actor_optimizer.step()
critic_optimizer.zero_grad()
value_loss=torch.stack(value_losses).sum()
value_loss.backward()
# print('loss: ', policy_loss, value_loss)
critic_optimizer.step()
del actor_net.saved_logprobs[:]
del critic_net.saved_values[:]
del critic_net.saved_nextvalues[:]
class NormalizedActions(gym.ActionWrapper): # gym env wrapper
def _action(self, action):
low = self.action_space.low
high = self.action_space.high
action = low + (action + 1.0) * 0.5 * (high - low)
action = np.clip(action, low, high)
return action
def _reverse_action(self, action):
low = self.action_space.low
high = self.action_space.high
action = 2 * (action - low) / (high - low) - 1
action = np.clip(action, low, high)
return action
def plot(frame_idx, rewards):
clear_output(True)
plt.figure(figsize=(20,5))
# plt.subplot(131)
plt.title('frame %s. reward: %s' % (frame_idx, rewards[-1]))
plt.plot(rewards)
# plt.plot(predict_qs)
plt.savefig('ac.png')
# plt.show()
ON_POLICY=True
hidden_dim = 30
UPDATE=['Approach0', 'Approach1'][0]
# choose env
ENV = ['Pendulum-v0', 'CartPole-v0', 'Reacher'][1] # Pendulum is continuous, CartPole is discrete
if ENV == 'Reacher':
DISCRETE = False
hidden_dim = 512
NUM_JOINTS=2
LINK_LENGTH=[200, 140]
INI_JOING_ANGLES=[0.1, 0.1]
# NUM_JOINTS=4
# LINK_LENGTH=[200, 140, 80, 50]
# INI_JOING_ANGLES=[0.1, 0.1, 0.1, 0.1]
SCREEN_SIZE=1000
SPARSE_REWARD=False
SCREEN_SHOT=False
action_range = 10.0
env=Reacher(screen_size=SCREEN_SIZE, num_joints=NUM_JOINTS, link_lengths = LINK_LENGTH, \
ini_joint_angles=INI_JOING_ANGLES, target_pos = [369,430], render=True, change_goal=False)
action_dim = env.num_actions
state_dim = env.num_observations
else: # gym env
if ENV == 'CartPole-v0':
DISCRETE = True
hidden_dim = 30
elif ENV == 'Pendulum-v0':
DISCRETE = False
hidden_dim = 128
if DISCRETE:
env = gym.make(ENV) # discrete env no normalizedactions
action_dim = env.action_space.n
else:
env = NormalizedActions(gym.make(ENV))
action_dim = env.action_space.shape[0]
state_dim = env.observation_space.shape[0]
action_range=1.
actor_net = ActorNetwork(state_dim, action_dim, hidden_dim).to(device)
critic_net = CriticNetwork(state_dim, hidden_dim).to(device)
print('Actor Network: ', actor_net)
print('Critic Network: ', critic_net)
actor_optimizer = optim.Adam(actor_net.parameters(), lr=1e-3)
critic_optimizer = optim.Adam(critic_net.parameters(), lr=1e-2)
def train():
# hyper-parameters
max_episodes = 3000
if ENV == 'Reacher':
max_steps = 20
elif ENV == 'Pendulum-v0':
max_steps = 150 # Pendulum needs 150 steps per episode to learn well, cannot handle 20
elif ENV == 'CartPole-v0':
max_steps = 1000 # short time step would be too easy for CartPole
frame_idx = 0
running_reward = 10
episode_rewards = []
# SavedTuple = namedtuple('SavedSet', ['log_prob', 'state_value'])
# SavedTuple2 = namedtuple('SavedSet2', ['log_prob', 'state_value', 'next_state_value'])
for i_episode in range (max_episodes):
if ENV == 'Reacher':
state = env.reset(SCREEN_SHOT)
# elif ENV == 'Pendulum':
else: # gym env
state = env.reset()
episode_reward = 0
if ON_POLICY:
rewards=[]
if not DETERMINISTIC:
entropies=0
for step in range (max_steps):
frame_idx+=1
if ON_POLICY:
action, log_prob, entropy = actor_net.evaluate_action(state)
# print('state: ', state, 'action: ', action, 'log_prob: ', log_prob)
state_value = critic_net(state)
if ENV == 'Reacher':
next_state, reward, done, _ = env.step(action, SPARSE_REWARD, SCREEN_SHOT)
# elif ENV == 'Pendulum':
else: # gym env
if DISCRETE:
next_state, reward, done, _ = env.step(action[0]) # discrete action only needs a index
else:
next_state, reward, done, _ = env.step(action)
env.render()
next_state_value = critic_net(next_state)
actor_net.saved_entropies.append(entropy)
if UPDATE == 'Approach0':
# this is critical! have to save the values inside the model to keep track of its gradients
actor_net.saved_logprobs.append(log_prob)
critic_net.saved_values.append(state_value)
# SavedSet.append(SavedTuple(log_prob, state_value))
if UPDATE == 'Approach1':
# this is critical! have to save the values inside the model to keep track of its gradients
actor_net.saved_logprobs.append(log_prob)
critic_net.saved_values.append(state_value)
critic_net.saved_nextvalues.append(next_state_value)
# SavedSet.append(SavedTuple2(log_prob, state_value, next_state_value))
if done:
reward = -20 if ENV == 'CartPole-v0' else reward
break
rewards.append(reward)
else: # off-policy update with memory buffer
pass
if done:
reward = -20 if ENV == 'CartPole-v0' else reward
break
state = next_state
episode_reward += reward
running_reward = running_reward * 0.99 + episode_reward * 0.01
# rewards.append(episode_reward)
if frame_idx%500==0:
plot(frame_idx, episode_rewards)
print('Episode: ', i_episode, '| Episode Reward: ', episode_reward, '| Running Reward: ', running_reward)
episode_rewards.append(episode_reward)
if UPDATE == 'Approach0':
Update0(rewards)
if UPDATE == 'Approach1':
Update1(rewards)
def main():
train()
if __name__ == '__main__':
main()