-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathutils.py
168 lines (154 loc) · 8.48 KB
/
utils.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
import os
import tqdm
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader, Dataset
import pandas as pd
from dataset import SlidingWindowDataset,SlidingWindowDataset_test
from models.transformer import TransformerModel,TransformerModel_NAT
import numpy as np
import logging
from scipy.stats import zscore
from losses import weighted_mse_loss
import matplotlib.pyplot as plt
def val(model,logger,model_path,training,plot_save_path ,if_save_plot=True,device=None):
#model.load_state_dict(torch.load('transformer_model.pth'))
if training:
model.eval()
else:
model.load_state_dict(torch.load(model_path))
model.to(device)
criterion_1 = nn.MSELoss()
criterion_2 = nn.L1Loss()
test_path = './dataset_test_v0'
test_dataset = SlidingWindowDataset(test_path, 30, None,None,is_training=False)
whole_result = []
if if_save_plot:
plot_result = os.path.join(plot_save_path, 'plot')
os.makedirs(plot_result, exist_ok=True)
for csv in tqdm.tqdm(os.listdir(test_path)[:50]):
mse_result = []
rmse_result = []
corre_result = []
# if '104060600017' in csv:
df = pd.read_csv(os.path.join(test_path, csv), index_col=0)
gt_y = df['y'].copy().tail(120).reset_index(drop=True)
test_df = df.tail(120).reset_index(drop=True)
dec_input = df['y'].tail(120).copy().reset_index(drop=True)
dec_input.loc[20:] = 0
for i in range(0,len(test_df) - 25 + 1,5):
stocks_feature, y = test_dataset.test_preprocess_v1(test_df[i:i+25])
src = torch.tensor(stocks_feature,dtype=torch.float32).unsqueeze(0).to(device)
src = src[:, :, :20]
final_out = []
for j in range(5):
if j==0:
out = torch.tensor(dec_input[i:i+20].values, dtype=torch.float32).unsqueeze(1).unsqueeze(1).to(device)
output = model.predict(src, out, day_num=29)
final_out.append(output[-1, :, :].item())
out = torch.cat((out,output[-1,:,:].view(1,1,1)),dim=0)
dec_input[i+20:i+20+5] = final_out
fig, ax = plt.subplots()
ax.plot([i for i in range(100)], dec_input[20:], label='predict', color='red')
ax.plot([i for i in range(100)], gt_y[20:], label='gt', color='blue')
ax.set_title(csv)
plt.savefig(os.path.join(plot_result,csv.replace('.csv','.jpg')),dpi=200)
predict = torch.tensor(np.array(dec_input[20:].values),dtype=torch.float32).unsqueeze(0).to(device)
gt_y = torch.tensor(np.array(gt_y[20:].values),dtype=torch.float32).unsqueeze(0).to(device)
mse_loss = criterion_1(predict, gt_y)
rmse_loss = torch.sqrt(mse_loss)
correlation_matrix = torch.corrcoef(torch.cat((predict, gt_y), dim=0))
corre_result.append(correlation_matrix[1,0].item())
mse_result.append(mse_loss.item())
rmse_result.append(rmse_loss.item())
logger.info('val mse loss : {:.4f}'.format(sum(mse_result)/len(mse_result)))
logger.info('val rmse loss : {:.4f}'.format(sum(rmse_result) / len(rmse_result)))
logger.info('correlation : {:.4f}'.format(sum(corre_result) / len(corre_result)))
whole_result.append(rmse_loss)
return sum(whole_result)/len(whole_result)
def val_NAT(model,logger,epoch,model_path,training,plot_save_path ,if_save_plot=True,device=None):
#model.load_state_dict(torch.load('transformer_model.pth'))
if training:
model.eval()
else:
model.load_state_dict(torch.load(model_path))
model.to(device)
criterion_1 = nn.MSELoss()
criterion_2 = nn.L1Loss()
test_path = './dataset_test_v0'
test_dataset = SlidingWindowDataset(test_path, 30, None,None,is_training=False)
whole_result = []
if if_save_plot:
plot_result = os.path.join(plot_save_path, 'plot_{}'.format(epoch))
os.makedirs(plot_result, exist_ok=True)
test_result = {
'mse_result':[],
'rmse_result':[],
'corre_result':[],
'mse_result_zscore':[],
'rmse_result_zscore':[],
'corre_result_zscore':[],
}
test_list = open('test_file.txt','r').readlines()
for csv in tqdm.tqdm(test_list[0:20]):
df = pd.read_csv(os.path.join(test_path, csv.strip()), index_col=0)
test_df = df.tail(119).reset_index(drop=True)
dec_input = df['y'].tail(119).copy().reset_index(drop=True)
dec_input.loc[19:] = 0
dec_input[0:19] = zscore(dec_input[0:19])
final_out = []
final_out_zscore = []
# dec_input[0:19] = zscore(dec_input[0:19])
# max_val = max(dec_input)
# min_val = min(dec_input)
# dec_input[0:19] = (dec_input[0:19]-min_val)/(max_val-min_val)
gt_y = []
for i in range(0,len(test_df) - 20 + 1,1):
stocks_feature, y = test_dataset.test_preprocess_v1(test_df[i:i+20])
src = torch.tensor(np.array(stocks_feature),dtype=torch.float32).unsqueeze(0).to(device) #N d L
tgt = torch.tensor(dec_input[i:i+19].values, dtype=torch.float32).unsqueeze(0).unsqueeze(0).to(device) # N d L
output = model.forward(src, tgt)
#print(output)
final_out.append(output.item())
# final_out_zscore.append(reg.item()/10)
dec_input[i+19] = output.item()
fig, ax = plt.subplots(1,1)
ax.plot([i for i in range(100)], np.array(final_out), label='predict', color='red')
ax.plot([i for i in range(100)], zscore(test_df['y'][19:]), label='y', color='blue')
# ax[1].plot([i for i in range(100)], test_df['y'][19:], label='y_zscore', color='green')
# ax[1].plot([i for i in range(100)], final_out_zscore, label='y', color='blue')
ax.set_title(csv)
ax.legend()
plt.savefig(os.path.join(plot_result,csv.strip().replace('.csv','.jpg')),dpi=200)
predict = torch.tensor(np.array(np.array(final_out)),dtype=torch.float32).unsqueeze(0).to(device)
gt_y = torch.tensor(test_df['y'][19:].values,dtype=torch.float32).unsqueeze(0).to(device)
# predict_zscore = torch.tensor(np.array(final_out_zscore),dtype=torch.float32).unsqueeze(0).to(device)
# gt_y_zscore = torch.tensor(np.array(test_df['y_zscore'][19:].values),dtype=torch.float32).unsqueeze(0).to(device)
#
# mse_loss_zscore = criterion_1(predict_zscore, gt_y_zscore)
# rmse_loss_zscore = torch.sqrt(mse_loss_zscore)
# correlation_matrix_zscore = torch.corrcoef(torch.cat((predict_zscore, gt_y_zscore), dim=0))
# test_result['corre_result_zscore'].append(correlation_matrix_zscore[1, 0].item())
# test_result['mse_result_zscore'].append(mse_loss_zscore.item())
# test_result['rmse_result_zscore'].append(rmse_loss_zscore.item())
mse_loss = criterion_1(predict, gt_y)
rmse_loss = torch.sqrt(mse_loss)
correlation_matrix = torch.corrcoef(torch.cat((predict, gt_y), dim=0))
test_result['corre_result'].append(correlation_matrix[1,0].item())
test_result['mse_result'].append(mse_loss.item())
test_result['rmse_result'].append(rmse_loss.item())
logger.info('val mse loss : {:.4f}'.format(sum(test_result['mse_result'])/len(test_result['mse_result'])))
logger.info('val rmse loss : {:.4f}'.format(sum(test_result['rmse_result']) / len(test_result['rmse_result'])))
logger.info('correlation : {:.4f}'.format(sum(test_result['corre_result']) / len(test_result['corre_result'])))
# logger.info('val mse zscore loss : {:.4f}'.format(sum(test_result['mse_result_zscore']) / len(test_result['mse_result_zscore'])))
# logger.info('val rmse zscore loss : {:.4f}'.format(sum(test_result['rmse_result_zscore']) / len(test_result['rmse_result_zscore'])))
# logger.info('correlation zscore: {:.4f}'.format(sum(test_result['corre_result_zscore']) / len(test_result['corre_result_zscore'])))
return sum(test_result['mse_result']) / len(test_result['mse_result'])
def scale_to_range(column):
min_val = np.min(column)
max_val = np.max(column)
return (column - min_val) / (max_val - min_val)
def inverse_scale_to_range(column,min_val,max_val):
column = ((column+1)/2)*(max_val - min_val)+min_val
return column