-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
123 lines (97 loc) · 4.04 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
import torch
import torch.nn as nn
from torch.nn import Embedding, Linear, MSELoss, CrossEntropyLoss, BCEWithLogitsLoss
from torch.nn.functional import relu,tanh,sigmoid
import pandas as pd
from dataset import SSBMDataset
from torch.utils.data import DataLoader
from tqdm import tqdm
from torch.optim import Adam
import pdb
from sklearn.metrics import precision_recall_fscore_support
import numpy as np
def top_n_accuracy(preds, targets, n):
best_n = torch.argsort(preds, dim=1)[:, -n:]
correct = 0
for i in range(targets.shape[0]):
if targets[i] in best_n[i,:]:
correct += 1
return correct / targets.shape[0]
def eval(model, val_dl, device, bce_crit):
total_mse_loss = 0
total_loss = 0
button_preds = []
button_targets = []
num_batch = 0
mse_crit = MSELoss(reduction='mean')
model.eval()
for batch in tqdm(val_dl):
num_batch += 1
features, cts_targets, bin_cls_targets = batch
with torch.no_grad():
cts_o, logits_o = model(features)
mse_loss = mse_crit(cts_o, cts_targets)
total_mse_loss += mse_loss
total_loss += bce_crit(logits_o, bin_cls_targets) + mse_loss
button_targets.append(bin_cls_targets)
button_preds.append(sigmoid(logits_o))
print(f'val_loss: {total_loss.item()/num_batch} val_mse: {total_mse_loss.item()/num_batch}')
button_preds = torch.cat(button_preds, dim=0).to('cpu').detach().numpy()
button_targets = torch.cat(button_targets, dim=0).to('cpu').detach().numpy()
report_button_cls_metrics(button_preds,button_targets)
def report_button_cls_metrics(preds, targets, thres=0.5):
# import pdb
# pdb.set_trace()
button_idx_to_name = {
0: "Y",
1: "X",
2: "B",
3: "A",
4: "L",
5: "R",
6: "Z"
}
preds = preds > thres
acc, recall, fscore, support = precision_recall_fscore_support(targets, preds)
for button_idx in range(preds.shape[1]):
if np.sum(targets[:, button_idx]) == 0:
continue
print(f'{button_idx_to_name[button_idx]} acc: {acc[button_idx]} recall: {recall[button_idx]} fscore: {fscore[button_idx]} support: {support[button_idx]}')
def train(model, trn_dl, val_dl, epoch, print_out_freq, device, pos_weigts):
model.to(device)
optim = Adam(model.parameters(), lr=0.0001)
mse_crit, bce_crit = MSELoss(reduction='mean'), BCEWithLogitsLoss(reduction='mean',pos_weight=torch.tensor(pos_weigts).to(device))
button_press_thres = 0.5
for i in range(epoch):
iter_num = 0
epoch_loss = 0
button_targets = []
button_preds = []
model.train()
for batch in tqdm(trn_dl):
iter_num += 1
optim.zero_grad()
features, cts_targets, bin_cls_targets = batch
# import pdb
# pdb.set_trace()
cts_o, logits_o = model(features)
mse_loss = mse_crit(cts_o, cts_targets)
# import pdb
# pdb.set_trace()
bce_loss = bce_crit(logits_o, bin_cls_targets)
total_loss = mse_loss + bce_loss
total_loss.backward()
epoch_loss += total_loss.item()
optim.step()
button_preds.append(sigmoid(logits_o))
button_targets.append(bin_cls_targets)
if iter_num % print_out_freq == 0:
print(f'epoch: {i} trn_loss: {epoch_loss / iter_num}')
button_preds = torch.cat(button_preds, dim=0).to('cpu').detach().numpy()
button_targets = torch.cat(button_targets, dim=0).to('cpu').detach().numpy()
print(f'end of {i}th epoch trn_loss: {epoch_loss / iter_num}')
report_button_cls_metrics(button_preds, button_targets)
print(f'Eval epoch {i}')
eval(model, val_dl, device, bce_crit)
# import pdb
# pdb.set_trace()