-
Notifications
You must be signed in to change notification settings - Fork 0
/
supervision.py
180 lines (145 loc) · 6.31 KB
/
supervision.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
from utils import *
from transformer import Transformer
import numpy as np
import os
import matplotlib.pyplot as plt
import time
import random
import json
import torch
from torch import nn
from torch.utils.data import ConcatDataset, SubsetRandomSampler, DataLoader
from torch.optim.lr_scheduler import ReduceLROnPlateau
import torch.nn.functional as f
def supervision(args):
train_type, train_K, train_N, train_T, train_Q, train_L = load_instance(args.train_index, 'train')
name = train_type + str(train_K) + '-' + str(train_N)
path_dataset = ['./dataset/' + file for file in os.listdir('./dataset') if file.startswith('dataset-' + name)]
training_sets = []
for file in path_dataset:
print('Load', file)
training_sets.append(torch.load(file))
os.remove(file)
training_set = ConcatDataset(training_sets)
data_size = len(training_set)
path_model = './model/'
os.makedirs(path_model, exist_ok=True)
path_result = './result/'
os.makedirs(path_result, exist_ok=True)
torch.manual_seed(0)
random.seed(0)
indices = list(range(data_size))
split = int(np.floor(0.02 * data_size))
train_indices, valid_indices = indices[split:], indices[:split]
print('The size of training set: {}.'.format(len(train_indices)),
'The size of validation set: {}.\n'.format(len(valid_indices)))
shuffle = True
if shuffle:
np.random.shuffle(train_indices)
np.random.shuffle(valid_indices)
train_sampler = SubsetRandomSampler(train_indices)
valid_sampler = SubsetRandomSampler(valid_indices)
train_data = DataLoader(training_set, batch_size=args.batch_size, sampler=train_sampler)
valid_data = DataLoader(training_set, batch_size=args.batch_size, sampler=valid_sampler)
# Determine if your system supports CUDA
cuda_available = torch.cuda.is_available()
device = get_device(cuda_available)
model = Transformer(
device=device,
num_vehicles=train_K,
input_seq_len=train_N,
target_seq_len=train_N + 2,
d_model=args.d_model,
num_layers=args.num_layers,
num_heads=args.num_heads,
d_k=args.d_k,
d_v=args.d_v,
d_ff=args.d_ff,
dropout=args.dropout)
model_name = name + '-' + str(args.wait_time)
if cuda_available:
model.cuda()
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
scheduler = ReduceLROnPlateau(optimizer, mode='min', patience=50, factor=0.99)
epochs = args.epochs
train_performance = np.zeros(epochs)
valid_performance = np.zeros(epochs)
exec_times = np.zeros(epochs)
model_validation = True
for epoch in range(epochs):
print('--------Training for Epoch {} starting:--------'.format(epoch))
start = time.time()
running_loss = 0
iters = 0
model.train()
for _, (states, actions) in enumerate(train_data):
iters += 1
actions = actions.to(device)
optimizer.zero_grad()
outputs = model(states)
loss = criterion(outputs, actions)
loss.backward()
optimizer.step()
running_loss += loss.item()
scheduler.step(running_loss)
_, predicted = torch.max(f.softmax(outputs, dim=1), 1)
train_measure = (predicted == actions).sum().item() / actions.size(0)
if iters % 20 == 0:
print('Epoch: {}.'.format(epoch),
'Iteration: {}.'.format(iters),
'Training loss: {:.4f}.'.format(loss.item()),
'Training accuracy: {:.4f}.'.format(train_measure)
)
train_performance[epoch] = train_performance[epoch] + train_measure
train_performance[epoch] /= (iters + 1)
if model_validation:
# Validation
print('-------Validation for Epoch {} starting:-------'.format(epoch))
model.eval()
with torch.no_grad():
valid_measure = [0, 0]
# Loop over batches in an epoch using valid_data
for _, (states, actions) in enumerate(valid_data):
actions = actions.to(device)
outputs = model(states)
loss = criterion(outputs, actions)
_, predicted = torch.max(f.softmax(outputs, dim=1), 1)
valid_measure[0] += (predicted == actions).sum().item()
valid_measure[1] += actions.size(0)
valid_performance[epoch] = valid_measure[0] / valid_measure[1]
print('Validation accuracy: {:.4f}.'.format(valid_performance[epoch]))
torch.save({
'epoch': epoch,
'model_state_dict': model.state_dict(),
'optimizer_state_dict': optimizer.state_dict(),
'scheduler_state_dict': scheduler.state_dict(),
'loss': criterion,
}, './model/' + 'sl-' + model_name + '.model')
end = time.time()
exec_time = end - start
exec_times[epoch] = exec_time
print('-> Execution time for Epoch {}: {:.4f} seconds.'.format(epoch, exec_time))
print('-> Estimated execution time remaining: {:.4f} seconds.\n'.format(exec_time * (epochs - epoch - 1)))
with open(path_result + 'training_log.txt', 'a+') as file:
json.dump({
'epoch': epoch,
'execution time': exec_time / 3600,
'estimated execution time remaining': exec_time * (epochs - epoch - 1) / 3600,
}, file)
file.write("\n")
print('Training finished.')
print('Average execution time per epoch: {:.4f} seconds.'.format(np.mean(exec_times)))
print("Total execution time: {:.4f} seconds.\n".format(np.sum(exec_times)))
fig, ax = plt.subplots()
file_name = 'accuracy-' + name + '-' + str(args.wait_time)
ax.plot(np.arange(epochs), train_performance, label="Training accuracy")
ax.plot(np.arange(epochs), valid_performance, label="Validation accuracy")
ax.set_xlabel('Epoch')
ax.set_ylabel('Accuracy')
ax.set(ylim=(0, 1))
ax.legend()
plt.savefig(path_result + file_name + '.pdf')
with open(path_result + file_name + '.npy', 'wb') as file:
np.save(file, train_performance) # noqa
np.save(file, valid_performance) # noqa