-
Notifications
You must be signed in to change notification settings - Fork 0
/
robust_train.py
273 lines (223 loc) · 9.31 KB
/
robust_train.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
import gym
import os, sys
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as f
from torch.autograd import Variable
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import time
import random
from logger import Logger
from memories import ExperienceReplay
from gym import wrappers
from gym.wrappers import AtariPreprocessing
import qnet_agentsSAC_auto
import json
import argparse
from tqdm import tqdm
# Set GPU ID
# os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID"
# os.environ["CUDA_VISIBLE_DEVICES"]= "7"
def moving_average(a, n=3) :
ret = np.cumsum(a, dtype=float)
ret[n:] = ret[n:] - ret[:-n]
return ret[n - 1:] / n
parser = argparse.ArgumentParser(description='Train SAC algorithm on Space Invaders')
parser.add_argument('--config', help="Json file with all the metaparameters. See config01.json as an example.", type=str, default="config01.json",dest="config_file")
parser.add_argument('--new', help="If 1 (default) the training starts from scracth, if 0 it starts from an old configuration.", type=int,default=True,dest="new_flag")
parser.add_argument('--game', type=str, default="BeamRider", dest="game")
args = parser.parse_args()
##############
#PARAMS
print("reading parameters...")
config_file = args.config_file
new_flag = bool(args.new_flag)
game = args.game
config = json.load(open(config_file))
#Id
configId = config["configId"]
#env
screen_size = config["env_parameters"]["screen_size"]
frame_skip = config["env_parameters"]["frame_skip"]
seed_value = config["env_parameters"]["seed_value"]
#agent
gamma = config["agent_parameters"]["gamma"]
lr_Q = config["agent_parameters"]["lr_Q"]
lr_pi = config["agent_parameters"]["lr_pi"]
lr_alpha = config["agent_parameters"]["lr_alpha"]
tau = config["agent_parameters"]["tau"]
h_dim = config["agent_parameters"]["h_dim"]
h_mu_dim = config["agent_parameters"]["h_mu_dim"]
alpha = config["agent_parameters"]["alpha"]
entropy_rate = config["agent_parameters"]["entropy_rate"]
#training
n_episodes = int(config["training_parameters"]["n_episodes"])
n_steps = int(config["training_parameters"]["n_steps"])
batch_size = config["training_parameters"]["batch_size"]
t_tot_cut = config["training_parameters"]["t_tot_cut"]
memory_size = int(config["training_parameters"]["memory_size"])
train_start = int(config["training_parameters"]["train_start"])
reg_train_start = int(config["training_parameters"]["reg_train_start"])
##############
#SETUP
print("setting up environment and agent...")
gameID = game + "-v4"
env = gym.make(gameID)
env.spec.id = gameID+"NoFrameskip"
print("action_meaning:", env.unwrapped.get_action_meanings())
env.seed(seed_value)
torch.manual_seed(seed_value)
np.random.seed(seed_value)
random.seed(seed_value)
env = wrappers.AtariPreprocessing(env,grayscale_obs=True,frame_skip=8,grayscale_newaxis=True,screen_size=screen_size)
n_states = env.observation_space.shape[0]
n_actions = env.action_space.n
QNet_Agent = qnet_agentsSAC_auto.QNet_Agent
memory = ExperienceReplay()
qnet_agent = QNet_Agent(n_states=n_states,
n_actions=n_actions,
gamma = gamma,
lr_Q = lr_Q,
lr_pi = lr_pi,
lr_alpha = lr_alpha,
tau = tau,
h_dim = h_dim,
h_mu_dim = h_mu_dim,
entropy_rate = entropy_rate,
alpha = alpha,
reg_start = reg_train_start
).cuda()
save_point = 0
if not new_flag:
print("reading old configuration from {}.json..".format(configId))
save_point = torch.load("./saved_models/INFO_{}_Q_SAC_auto_{}.model".format(game, configId))
qnet_agent.Q.load_state_dict(torch.load("./saved_models/{}_Q_SAC_auto_{}.model".format(game, configId)))
qnet_agent.target_Q.load_state_dict(torch.load("./saved_models/{}_target_Q_SAC_auto_{}.model".format(game, configId)))
qnet_agent.pi.load_state_dict(torch.load("./saved_models/{}_pi_SAC_auto_{}.model".format(game, configId)))
########################################################
# test pretrained Models
def test():
print("test Mode")
state = env.reset()
state = np.transpose(state, [2,0,1])
episode_steps = 0
episode_return = 0.0
f_lives = env.unwrapped.ale.lives()
done = False
step=0
while True:
try:
state_cuda = torch.Tensor(state).unsqueeze(0)
action = qnet_agent.exploit_action(state_cuda)
if step==0:
action = 2
new_state, reward, done, info = env.step(action)
lives = env.unwrapped.ale.lives()
if lives < f_lives:
f_lives = lives
env.step(2)
###
episode_return += reward
new_state = np.transpose(new_state, [2,0,1])
state = new_state
step+=1
if done or step>1e4:
break
except KeyboardInterrupt:
env.close()
print("break")
break
#########################
print('episode Reward: ' + str(episode_return))
return episode_return
########################################################
########################################################
print("start training...")
rewards_per_episode = []; time_per_episode = []; t_total = save_point
time_start = time.time()
best_episode_reward = -float("inf")
######
logger = Logger()
####
tot_steps = n_steps
p_bar = tqdm(total=tot_steps)
for i_episode in range(len(rewards_per_episode),n_episodes):
logger.on()
env = gym.make(gameID)
env.spec.id = gameID+"NoFrameskip"
env = wrappers.AtariPreprocessing(env,grayscale_obs=True,frame_skip=frame_skip,grayscale_newaxis=True,screen_size=screen_size)
try:
state = env.reset()
state = np.transpose(state, [2,0,1])
t=0
rewards = []
Q_loss, pi_loss, entropy_loss, action_reg_loss = 0, 0 ,0, 0
while True:
t+=1
t_total+=1
state_cuda = torch.Tensor(state).unsqueeze(0)
action = qnet_agent.select_action(state_cuda)
new_state, reward, done, info = env.step(action)
new_state = np.transpose(new_state, [2,0,1])
memory.push(state, action, new_state, reward, done)
if memory.__len__()>=train_start and t%4==0:
batch = memory.sample(batch_size)
Q_loss, pi_loss, entropy_loss, action_reg_loss= qnet_agent.optimize(batch, t_total)
Q_loss = Q_loss.detach().item()
pi_loss = pi_loss.detach().item()
entropy_loss = entropy_loss.detach().item()
if t_total > reg_train_start:
action_reg_loss = action_reg_loss.detach().item()
state = new_state
rewards.append(reward)
if t>=t_tot_cut: break
if done: break
tot_rewards = np.sum(rewards)
rewards_per_episode.append(tot_rewards)
best_episode_reward = max(rewards_per_episode)
time_per_episode.append(t_total)
eval_reward= -1
if i_episode % 20 == 0:
eval_reward = test()
if i_episode%1==0:
p_bar.update(t_total -p_bar.n)
if t_total >= tot_steps:
p_bar.close()
sys.exit(0)
logger.off()
logger.log(i_episode, tot_rewards, entropy_loss, Q_loss, pi_loss, t_total, eval_reward, action_reg_loss)
logger.on()
alpha = qnet_agent.alpha.detach().item()
elapsed_time = round(time.time() - time_start)
dbin = 1000
s = "i_episode = {0}, t_total = {1}".format(i_episode, t_total)
s += "\nlast avg. reward = {0}, elapsed_time = {1} sec.".format(np.mean(rewards_per_episode[-dbin:]),elapsed_time)
s+= "\nalpha = {0}".format(qnet_agent.alpha.cpu().item())
s+= "\nbest episode reward = {}".format(best_episode_reward)
print("training {}".format(configId))
print(s)
print("action_reg_loss",action_reg_loss)
print("pi_loss",pi_loss)
print()
if i_episode%100==0:
plt.title(s)
plt.plot(rewards_per_episode,alpha=0.4)
plt.hlines(0,0,len(rewards_per_episode))
if len(rewards_per_episode)>dbin*2:
ma_rewards_per_episode = moving_average(rewards_per_episode,dbin)
plt.plot(ma_rewards_per_episode)
plt.tight_layout()
plt.savefig("./train_figs/train_curve_{}.png".format(configId), dpi=300)
plt.close()
if i_episode%10==0:
torch.save(t_total, "./saved_models/INFO_{}_Q_SAC_auto_{}.model".format(game, configId))
torch.save(qnet_agent.Q.state_dict(), "./saved_models/{}_Q_SAC_auto_{}.model".format(game, configId))
torch.save(qnet_agent.target_Q.state_dict(), "./saved_models/{}_target_Q_SAC_auto_{}.model".format(game, configId))
torch.save(qnet_agent.pi.state_dict(), "./saved_models/{}_pi_SAC_auto_{}.model".format(game, configId))
torch.cuda.empty_cache()
except KeyboardInterrupt:
print("break")
break