/
train.py
168 lines (139 loc) · 4.97 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
from model import *
from utils import binMolDen
import matplotlib.pyplot as plt
import os
import numpy as np
import pandas as pd
from random import sample
from sklearn import metrics
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.autograd import Variable
from torch.optim.lr_scheduler import CosineAnnealingLR, StepLR
from torch.utils.data import dataset, DataLoader
# writer = SummaryWriter(log_dir='train/log/toy')
# data = pd.read_csv('data/train.csv', header=None, index_col=None).to_numpy()
data = np.load('data/train.npz')['data']
train_data, valid_data= split_data(
data, valid_ratio = 0.00,
randomSeed = 2022
)
test_data = np.load('data/test.npz')['data']
print(len(test_data))
print('Test size =', len(test_data))
train_dataset = IonDataset(data = train_data)
valid_dataset = IonDataset(data = valid_data)
test_dataset = IonDataset(data = test_data)
train_loader = DataLoader(
train_dataset, batch_size=1024, num_workers=0, drop_last=False,
shuffle=True
)
valid_loader = DataLoader(
valid_dataset, batch_size=1024, num_workers=0, drop_last=False,
shuffle=False
)
test_loader = DataLoader(
test_dataset, batch_size=1024, num_workers=0, drop_last=False,
shuffle=False
)
criterion = nn.MSELoss()
if torch.cuda.is_available():
device = 'cuda'
else:
device = 'cpu'
model = IonNet(n_in=6, activation='ReLU').to(device)
# optimizer = optim.Adam(model.parameters(), lr=0.005)
optimizer = optim.Adam(model.parameters(), lr=0.005)
scheduler = StepLR(optimizer, step_size=1000, gamma=0.1)
model_checkpoints_folder = os.path.join('train/', 'NN')
#n_epochs=100
n_epochs=2
log_every = 500
# log_every = len(train_loader)
n_iter = 0
valid_n_iter = 0
best_valid_loss = np.inf
best_valid_mae = np.inf
best_valid_roc_auc = 0
for epoch_counter in range(n_epochs):
model.train()
for bn, (inputs, target) in enumerate(train_loader):
input_var = inputs.to(device)
target = target.to(device)
# compute output
optimizer.zero_grad()
output = model(input_var)
loss = criterion(output, target)
if bn % log_every == 0:
# writer.add_scalar('train_loss', loss.item(), global_step=n_iter)
print('Epoch: %d, Batch: %d, Loss:'%(epoch_counter+1, bn), loss.item())
loss.backward()
optimizer.step()
n_iter += 1
scheduler.step()
# validate the model if requested
if epoch_counter % 1 == 0:
losses = AverageMeter()
mae_errors = AverageMeter()
with torch.no_grad():
model.eval()
for bn, (inputs, target) in enumerate(valid_loader):
input_var = inputs.to(device)
target = target.to(device)
# compute output
output = model(input_var)
loss = criterion(output, target)
mae_error = mae(output, target)
mae_errors.update(mae_error, target.size(0))
print('Epoch [{0}] Validate: [{1}/{2}], '
'MAE {mae_errors.val:.3f} ({mae_errors.avg:.3f})'.format(
epoch_counter+1, bn+1, len(valid_loader),
mae_errors=mae_errors))
if mae_errors.avg < best_valid_mae:
# save the model weights
best_valid_mae = mae_errors.avg
torch.save(model.state_dict(), os.path.join(model_checkpoints_folder, 'model.pth'))
# writer.add_scalar('valid_loss', losses.avg, global_step=valid_n_iter)
valid_n_iter += 1
state_dict = torch.load(os.path.join(model_checkpoints_folder, 'model.pth'), map_location=device)
model.load_state_dict(state_dict)
mae_errors = AverageMeter()
with torch.no_grad():
model.eval()
for bn, (inputs, target) in enumerate(test_loader):
input_var = inputs.to(device)
target = target.to(device)
# compute output
output = model(input_var)
mae_error = mae(output.data, target)
mae_errors.update(mae_error, target.size(0))
print('Epoch [{0}] Test: [{1}/{2}], '
'MAE: {mae_errors.avg:.3f}'.format(
epoch_counter+1, bn+1, len(test_loader),
mae_errors=mae_errors))
# bins = np.linspace(0, 1.9+0.2, 100)
gap=2.0
binSize=0.02
bins = np.array([i*binSize for i in range(int((gap/2+0.1)//binSize))])
with torch.no_grad():
model.eval()
cdf = []
for b in bins:
p = model(torch.tensor([b, gap, 2.2, 4.8305, 0.0128, -1]).float().to(device)).cpu().numpy()
cdf.append(p)
cdf = np.array(cdf)
pdf = cdf[1:]-cdf[:-1]
# conc_profile = binMolDen(3.1928623, 3.4032, binSize, pdf*len(ions)/2)
fig, ax = plt.subplots(figsize = (6, 3), dpi=600)
ax2 = ax.twinx()
ax.plot(bins, cdf, color = 'C0')
ax2.plot(bins[:-1], pdf, color = 'C1')
# giving labels to the axises
ax.set_xlabel('Distance from channel center ($\AA$)', fontsize=16)
ax.set_ylabel('CDF', color = 'C0', fontsize=16)
ax2.set_ylabel('PDF', color = 'C1', fontsize=16)
# defining display layout
plt.tight_layout()
plt.show()