-
Notifications
You must be signed in to change notification settings - Fork 0
/
model_train.py
249 lines (227 loc) · 9.89 KB
/
model_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
import pandas as pd
import numpy as np
from models import TimeDataSet, LSTMModel, LSTMDiseaseModel
from datetime import datetime
import torch
from tqdm import trange
import torch.nn as nn
import os
device = 'cuda' if torch.cuda.is_available() else 'cpu'
from utils import draw_arrays, draw_array
def date_to_integer(date):
base_date = datetime(date.year, 1, 1)
days_diff = (date - base_date).days
return days_diff
def get_time_dataset():
df = pd.read_excel("processed_data/time_sale_all.xlsx")
df.rename(columns={df.columns[0]: 'time'}, inplace=True)
df['time'] = df['time'].apply(date_to_integer)
time_volume = np.array(df).astype(float)
cost_price = pd.read_excel("processed_data/cost_all.xlsx")
cost_price = np.array(cost_price)[:, 1:].astype(float)
cost = cost_price[:, range(0, 11, 2)]
price = cost_price[:, range(1, 12, 2)]
data = np.concatenate([time_volume, cost, price], axis=1)
dataset = TimeDataSet(data)
return dataset
def train_lstm_model(epochs=100, step=30, mode="part", save=False, save_loss=False):
"""
train the lstm-ARIMA model
"""
# model & loss & optim
model = LSTMModel()
model.fit(get_time_dataset())
loss_func = nn.MSELoss().to(device)
optimizer = torch.optim.Adam(model.parameters())
filename = "pretrained_weight/lstm_weight_step_{}.pt".format(step)
if os.path.exists(filename):
model.load_state_dict(torch.load(filename))
# dataset and epochs
if mode == "part":
train_dataset = model.get_dataset("train", step)
valid_dataset = model.get_dataset("valid", step)
test_dataset = model.get_dataset("test", step)
best_val_score = 10000
test_loss_score = 0
elif mode == "all":
train_dataset = model.get_dataset("all", step)
elif mode == "last":
train_dataset = model.get_dataset("all", step)[-100:]
model.train()
train_epochs = trange(epochs, ascii=True, leave=True, desc="Epoch", position=0)
loss_list = list()
# train & valid
for epoch in train_epochs:
loss_score = 0
pred_diff_list = list()
y_diff_list = list()
for data in train_dataset:
x, x_corr, t, x_diff, _, y_diff = data
x = x.to(device)
x_corr = x_corr.to(device)
t = t.to(device)
x_diff = x_diff.to(device)
y_diff = y_diff.to(device)
optimizer.zero_grad()
pred_diff = model(x, x_corr, t, x_diff)
loss = loss_func(pred_diff, y_diff)
pred_diff_list.append(pred_diff.cpu().detach().numpy())
y_diff_list.append(y_diff.cpu().detach().numpy())
loss.backward()
loss_score += loss.item()
optimizer.step()
loss_score /= len(train_dataset)
train_epochs.set_description("Epoch (Loss=%g)" % round(loss_score, 5))
loss_list.append(loss_score)
if mode == "all" and epoch == epochs-1:
np_pred_diff = np.array(pred_diff_list)[:, 0][-100:]
np_y_diff = np.array(y_diff_list)[:, 0][-100:]
draw_arrays(np_pred_diff, np_y_diff, label1="pred", label2="fact",
save=True, filename="all_pred_lines_volume.png".format(epoch))
# valid
if mode == "part" and epoch % 10 == 9:
val_loss_score = 0
for data in valid_dataset:
x, x_corr, t, x_diff, _, y_diff = data
x = x.to(device)
x_corr = x_corr.to(device)
t = t.to(device)
x_diff = x_diff.to(device)
y_diff = y_diff.to(device)
pred_diff = model(x, x_corr, t, x_diff)
loss = loss_func(pred_diff, y_diff)
val_loss_score += loss.item()
val_loss_score /= len(valid_dataset)
print("\nvalid loss score: ", val_loss_score)
if val_loss_score < best_val_score:
best_val_score = val_loss_score
if save:
torch.save(model.state_dict(), filename)
np_pred_diff = np.array(pred_diff_list)[:, 0][-100:]
np_y_diff = np.array(y_diff_list)[:, 0][-100:]
draw_arrays(np_pred_diff, np_y_diff, label1="pred", label2="fact",
save=True, filename="epoch_{}_pred_lines_volume.png".format(epoch))
if save_loss:
torch.save(torch.Tensor(loss_list), "pretrained_weight/train_loss.pt")
draw_array(np.array(loss), save=True, filename="train_loss.png",
label="loss", x_label="Epoch", y_label="MSE loss",
title="lstm-ARIMA Model Train loss",
noise_method=False)
if save and (mode == "all" or mode == "last"):
torch.save(model.state_dict(), filename)
elif mode == "part":
# test
for data in test_dataset:
x, x_corr, t, x_diff, _, y_diff = data
x = x.to(device)
x_corr = x_corr.to(device)
t = t.to(device)
x_diff = x_diff.to(device)
y_diff = y_diff.to(device)
pred_diff = model(x, x_corr, t, x_diff)
loss = loss_func(pred_diff, y_diff)
test_loss_score += loss.item()
test_loss_score /= len(test_dataset)
print("test loss score: ", test_loss_score)
def train_lstm_model_disease(epochs=100, step=30, mode="part", save=False, save_loss=False):
"""
Train the LSTM-ARIMA Model Considering Disease
"""
# model & loss & optim
model = LSTMDiseaseModel()
model.fit(get_time_dataset())
loss_func = nn.MSELoss().to(device)
optimizer = torch.optim.Adam(model.parameters())
filename = "pretrained_weight/disease_lstm_weight_step_{}.pt".format(step)
if os.path.exists(filename):
model.load_state_dict(torch.load(filename))
# dataset and epochs
if mode == "part":
train_dataset = model.get_dataset("train", step)
valid_dataset = model.get_dataset("valid", step)
test_dataset = model.get_dataset("test", step)
best_val_score = 10000
test_loss_score = 0
elif mode == "all":
train_dataset = model.get_dataset("all", step)
elif mode == "last":
train_dataset = model.get_dataset("all", step)[-100:]
model.train()
train_epochs = trange(epochs, ascii=True, leave=True, desc="Epoch", position=0)
loss_list = list()
# train & valid
for epoch in train_epochs:
loss_score = 0
pred_diff_list = list()
y_diff_list = list()
for data in train_dataset:
x, x_corr, t, x_diff, _, y_diff, disease = data
x = x.to(device)
x_corr = x_corr.to(device)
t = t.to(device)
x_diff = x_diff.to(device)
y_diff = y_diff.to(device)
disease = disease.to(device)
optimizer.zero_grad()
pred_diff = model(x, x_corr, t, x_diff, disease)
loss = loss_func(pred_diff, y_diff)
pred_diff_list.append(pred_diff.cpu().detach().numpy())
y_diff_list.append(y_diff.cpu().detach().numpy())
loss.backward()
loss_score += loss.item()
optimizer.step()
loss_score /= len(train_dataset)
train_epochs.set_description("Epoch (Loss=%g)" % round(loss_score, 5))
loss_list.append(loss_score)
if mode == "all" and epoch == epochs-1:
np_pred_diff = np.array(pred_diff_list)[:, 0][-100:]
np_y_diff = np.array(y_diff_list)[:, 0][-100:]
draw_arrays(np_pred_diff, np_y_diff, label1="pred", label2="fact",
save=True, filename="disease_all_pred_lines_volume.png".format(epoch))
# valid
if mode == "part" and epoch % 10 == 9:
val_loss_score = 0
for data in valid_dataset:
x, x_corr, t, x_diff, _, y_diff, disease = data
x = x.to(device)
x_corr = x_corr.to(device)
t = t.to(device)
x_diff = x_diff.to(device)
y_diff = y_diff.to(device)
disease = disease.to(device)
pred_diff = model(x, x_corr, t, x_diff, disease)
loss = loss_func(pred_diff, y_diff)
val_loss_score += loss.item()
val_loss_score /= len(valid_dataset)
print("\nvalid loss score: ", val_loss_score)
if val_loss_score < best_val_score:
best_val_score = val_loss_score
if save:
torch.save(model.state_dict(), filename)
np_pred_diff = np.array(pred_diff_list)[:, 0][-100:]
np_y_diff = np.array(y_diff_list)[:, 0][-100:]
draw_arrays(np_pred_diff, np_y_diff, label1="pred", label2="fact",
save=True, filename="disease_epoch_{}_pred_lines_volume.png".format(epoch))
if save_loss:
torch.save(torch.Tensor(loss_list), "pretrained_weight/disease_train_loss.pt")
draw_array(np.array(loss), save=True, filename="disease_train_loss.png",
label="loss", x_label="Epoch", y_label="MSE loss",
title="lstm-ARIMA Disease Model Train loss",
noise_method=False)
if save and (mode == "all" or mode == "last"):
torch.save(model.state_dict(), filename)
elif mode == "part":
# test
for data in test_dataset:
x, x_corr, t, x_diff, _, y_diff, disease = data
x = x.to(device)
x_corr = x_corr.to(device)
t = t.to(device)
x_diff = x_diff.to(device)
y_diff = y_diff.to(device)
disease = disease.to(device)
pred_diff = model(x, x_corr, t, x_diff, disease)
loss = loss_func(pred_diff, y_diff)
test_loss_score += loss.item()
test_loss_score /= len(test_dataset)
print("test loss score: ", test_loss_score)