-
Notifications
You must be signed in to change notification settings - Fork 61
/
Copy pathtrain.py
171 lines (150 loc) · 6.99 KB
/
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
import copy
import json
import os
import random
import time
from collections import deque
import gym
import pandas as pd
import torch
import numpy as np
from visdom import Visdom
import PPO_model
from env.case_generator import CaseGenerator
from validate import validate, get_validate_env
def setup_seed(seed):
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
random.seed(seed)
torch.backends.cudnn.deterministic = True
def main():
# PyTorch initialization
# gpu_tracker = MemTracker() # Used to monitor memory (of gpu)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
if device.type == 'cuda':
torch.cuda.set_device(device)
torch.set_default_tensor_type('torch.cuda.FloatTensor')
else:
torch.set_default_tensor_type('torch.FloatTensor')
print("PyTorch device: ", device.type)
torch.set_printoptions(precision=None, threshold=np.inf, edgeitems=None, linewidth=None, profile=None, sci_mode=False)
# Load config and init objects
with open("./config.json", 'r') as load_f:
load_dict = json.load(load_f)
env_paras = load_dict["env_paras"]
model_paras = load_dict["model_paras"]
train_paras = load_dict["train_paras"]
env_paras["device"] = device
model_paras["device"] = device
env_valid_paras = copy.deepcopy(env_paras)
env_valid_paras["batch_size"] = env_paras["valid_batch_size"]
model_paras["actor_in_dim"] = model_paras["out_size_ma"] * 2 + model_paras["out_size_ope"] * 2
model_paras["critic_in_dim"] = model_paras["out_size_ma"] + model_paras["out_size_ope"]
num_jobs = env_paras["num_jobs"]
num_mas = env_paras["num_mas"]
opes_per_job_min = int(num_mas * 0.8)
opes_per_job_max = int(num_mas * 1.2)
memories = PPO_model.Memory()
model = PPO_model.PPO(model_paras, train_paras, num_envs=env_paras["batch_size"])
env_valid = get_validate_env(env_valid_paras) # Create an environment for validation
maxlen = 1 # Save the best model
best_models = deque()
makespan_best = float('inf')
# Use visdom to visualize the training process
is_viz = train_paras["viz"]
if is_viz:
viz = Visdom(env=train_paras["viz_name"])
# Generate data files and fill in the header
str_time = time.strftime("%Y%m%d_%H%M%S", time.localtime(time.time()))
save_path = './save/train_{0}'.format(str_time)
os.makedirs(save_path)
# Training curve storage path (average of validation set)
writer_ave = pd.ExcelWriter('{0}/training_ave_{1}.xlsx'.format(save_path, str_time))
# Training curve storage path (value of each validating instance)
writer_100 = pd.ExcelWriter('{0}/training_100_{1}.xlsx'.format(save_path, str_time))
valid_results = []
valid_results_100 = []
data_file = pd.DataFrame(np.arange(10, 1010, 10), columns=["iterations"])
data_file.to_excel(writer_ave, sheet_name='Sheet1', index=False)
writer_ave.save()
writer_ave.close()
data_file = pd.DataFrame(np.arange(10, 1010, 10), columns=["iterations"])
data_file.to_excel(writer_100, sheet_name='Sheet1', index=False)
writer_100.save()
writer_100.close()
# Start training iteration
start_time = time.time()
env = None
for i in range(1, train_paras["max_iterations"]+1):
# Replace training instances every x iteration (x = 20 in paper)
if (i - 1) % train_paras["parallel_iter"] == 0:
# \mathcal{B} instances use consistent operations to speed up training
nums_ope = [random.randint(opes_per_job_min, opes_per_job_max) for _ in range(num_jobs)]
case = CaseGenerator(num_jobs, num_mas, opes_per_job_min, opes_per_job_max, nums_ope=nums_ope)
env = gym.make('fjsp-v0', case=case, env_paras=env_paras)
print('num_job: ', num_jobs, '\tnum_mas: ', num_mas, '\tnum_opes: ', sum(nums_ope))
# Get state and completion signal
state = env.state
done = False
dones = env.done_batch
last_time = time.time()
# Schedule in parallel
while ~done:
with torch.no_grad():
actions = model.policy_old.act(state, memories, dones)
state, rewards, dones = env.step(actions)
done = dones.all()
memories.rewards.append(rewards)
memories.is_terminals.append(dones)
# gpu_tracker.track() # Used to monitor memory (of gpu)
print("spend_time: ", time.time()-last_time)
# Verify the solution
gantt_result = env.validate_gantt()[0]
if not gantt_result:
print("Scheduling Error！！！！！！")
# print("Scheduling Finish")
env.reset()
# if iter mod x = 0 then update the policy (x = 1 in paper)
if i % train_paras["update_timestep"] == 0:
loss, reward = model.update(memories, env_paras, train_paras)
print("reward: ", '%.3f' % reward, "; loss: ", '%.3f' % loss)
memories.clear_memory()
if is_viz:
viz.line(X=np.array([i]), Y=np.array([reward]),
win='window{}'.format(0), update='append', opts=dict(title='reward of envs'))
viz.line(X=np.array([i]), Y=np.array([loss]),
win='window{}'.format(1), update='append', opts=dict(title='loss of envs')) # deprecated
# if iter mod x = 0 then validate the policy (x = 10 in paper)
if i % train_paras["save_timestep"] == 0:
print('\nStart validating')
# Record the average results and the results on each instance
vali_result, vali_result_100 = validate(env_valid_paras, env_valid, model.policy_old)
valid_results.append(vali_result.item())
valid_results_100.append(vali_result_100)
# Save the best model
if vali_result < makespan_best:
makespan_best = vali_result
if len(best_models) == maxlen:
delete_file = best_models.popleft()
os.remove(delete_file)
save_file = '{0}/save_best_{1}_{2}_{3}.pt'.format(save_path, num_jobs, num_mas, i)
best_models.append(save_file)
torch.save(model.policy.state_dict(), save_file)
if is_viz:
viz.line(
X=np.array([i]), Y=np.array([vali_result.item()]),
win='window{}'.format(2), update='append', opts=dict(title='makespan of valid'))
# Save the data of training curve to files
data = pd.DataFrame(np.array(valid_results).transpose(), columns=["res"])
data.to_excel(writer_ave, sheet_name='Sheet1', index=False, startcol=1)
writer_ave.save()
writer_ave.close()
column = [i_col for i_col in range(100)]
data = pd.DataFrame(np.array(torch.stack(valid_results_100, dim=0).to('cpu')), columns=column)
data.to_excel(writer_100, sheet_name='Sheet1', index=False, startcol=1)
writer_100.save()
writer_100.close()
print("total_time: ", time.time() - start_time)
if __name__ == '__main__':
main()