Newer
Older
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
250
251
252
253
import os
import torch
import torch.optim as optim
import wandb as wdb
import common_dataset
import dataloader
from config_common import load_args
from common_dataset import load_data
from dataloader import load_data
from loss import masked_cos_sim, distance, masked_spectral_angle
from model_custom import Model_Common_Transformer, Model_Common_Transformer_TAPE
from model import RT_pred_model_self_attention_multi
def train(model, data_train, epoch, optimizer, criterion_rt, criterion_intensity, metric_rt, metric_intensity, forward,
wandb=None):
losses_rt = 0.
losses_int = 0.
dist_rt_acc = 0.
dist_int_acc = 0.
model.train()
for param in model.parameters():
param.requires_grad = True
if forward == 'both':
for seq, charge, rt, intensity in data_train:
rt, intensity = rt.float(), intensity.float()
if torch.cuda.is_available():
seq, charge, rt, intensity = seq.cuda(), charge.cuda(), rt.cuda(), intensity.cuda()
pred_rt, pred_int = model.forward(seq, charge)
loss_rt = criterion_rt(rt, pred_rt)
loss_int = criterion_intensity(intensity, pred_int)
loss = loss_rt + loss_int
dist_rt = metric_rt(rt, pred_rt)
dist_int = metric_intensity(intensity, pred_int)
dist_rt_acc += dist_rt.item()
dist_int_acc += dist_int.item()
losses_rt += loss_rt.item()
losses_int += 5.*loss_int.item()
optimizer.zero_grad()
loss.backward()
optimizer.step()
if wandb is not None:
wdb.log({"train rt loss": losses_rt / len(data_train), "train int loss": losses_int / len(data_train),
"train rt mean metric": dist_rt_acc / len(data_train),
"train int mean metric": dist_int_acc / len(data_train),
'train epoch': epoch})
print('epoch : ', epoch, 'train rt loss', losses_rt / len(data_train), 'train int loss',
losses_int / len(data_train), "train rt mean metric : ", dist_rt_acc / len(data_train),
"train int mean metric",
dist_int_acc / len(data_train))
if forward == 'rt':
for seq, rt in data_train:
rt = rt.float()
if torch.cuda.is_available():
seq, rt = seq.cuda(), rt.cuda()
pred_rt = model.forward_rt(seq)
loss_rt = criterion_rt(rt, pred_rt)
loss = loss_rt
dist_rt = metric_rt(rt, pred_rt)
dist_rt_acc += dist_rt.item()
losses_rt += loss_rt.item()
optimizer.zero_grad()
loss.backward()
optimizer.step()
if wandb is not None:
wdb.log({"train rt loss": losses_rt / len(data_train),
"train rt mean metric": dist_rt_acc / len(data_train),
'train epoch': epoch})
print('epoch : ', epoch, 'train rt loss', losses_rt / len(data_train), "train rt mean metric : ",
dist_rt_acc / len(data_train))
if forward == 'int':
for seq, charge, intensity in data_train:
intensity = intensity.float()
if torch.cuda.is_available():
seq, charge, intensity = seq.cuda(), charge.cuda(), intensity.cuda()
pred_int = model.forward_int(seq, charge)
loss_int = criterion_intensity(intensity, pred_int)
loss = loss_int
dist_int = metric_intensity(intensity, pred_int)
dist_int_acc += dist_int.item()
losses_int += loss_int.item()
optimizer.zero_grad()
loss.backward()
optimizer.step()
if wandb is not None:
wdb.log({"train int loss": losses_int / len(data_train),
"train int mean metric": dist_int_acc / len(data_train),
'train epoch': epoch})
print('epoch : ', epoch, 'train int loss',
losses_int / len(data_train),
"train int mean metric",
dist_int_acc / len(data_train))
def eval(model, data_val, epoch, criterion_rt, criterion_intensity, metric_rt, metric_intensity, forward, wandb=None):
losses_rt = 0.
losses_int = 0.
dist_rt_acc = 0.
dist_int_acc = 0.
for param in model.parameters():
param.requires_grad = False
if forward == 'both':
for seq, charge, rt, intensity in data_val:
rt, intensity = rt.float(), intensity.float()
if torch.cuda.is_available():
seq, charge, rt, intensity = seq.cuda(), charge.cuda(), rt.cuda(), intensity.cuda()
pred_rt, pred_int = model.forward(seq, charge)
loss_rt = criterion_rt(rt, pred_rt)
loss_int = criterion_intensity(intensity, pred_int)
losses_rt += loss_rt.item()
losses_int += loss_int.item()
dist_rt = metric_rt(rt, pred_rt)
dist_int = metric_intensity(intensity, pred_int)
dist_rt_acc += dist_rt.item()
dist_int_acc += dist_int.item()
if wandb is not None:
wdb.log({"val rt loss": losses_rt / len(data_val), "val int loss": losses_int / len(data_val),
"val rt mean metric": dist_rt_acc / len(data_val),
"val int mean metric": dist_int_acc / len(data_val),
'val epoch': epoch})
print('epoch : ', epoch, 'val rt loss', losses_rt / len(data_val), 'val int loss', losses_int / len(data_val),
"val rt mean metric : ",
dist_rt_acc / len(data_val), "val int mean metric", dist_int_acc / len(data_val))
if forward == 'rt': #adapted to prosit dataset format
for seq, rt in data_val:
rt = rt.float()
if torch.cuda.is_available():
seq, rt = seq.cuda(), rt.cuda()
pred_rt = model.forward_rt(seq)
loss_rt = criterion_rt(rt, pred_rt)
losses_rt += loss_rt.item()
dist_rt = metric_rt(rt, pred_rt)
dist_rt_acc += dist_rt.item()
if wandb is not None:
wdb.log({"val rt loss": losses_rt / len(data_val),
"val rt mean metric": dist_rt_acc / len(data_val),
'val epoch': epoch})
print('epoch : ', epoch, 'val rt loss', losses_rt / len(data_val),
"val rt mean metric : ",
dist_rt_acc / len(data_val))
if forward == 'int': #adapted to prosit dataset format
for seq, charge, _, intensity in data_val:
intensity = intensity.float()
if torch.cuda.is_available():
seq, charge, intensity = seq.cuda(), charge.cuda(), intensity.cuda()
pred_int = model.forward_int(seq, charge)
loss_int = criterion_intensity(intensity, pred_int)
losses_int += loss_int.item()
dist_int = metric_intensity(intensity, pred_int)
dist_int_acc += dist_int.item()
if wandb is not None:
wdb.log({"val int loss": losses_int / len(data_val),
"val int mean metric": dist_int_acc / len(data_val),
'val epoch': epoch})
print('epoch : ', epoch, 'val int loss', losses_int / len(data_val), "val int mean metric",
dist_int_acc / len(data_val))
def run(epochs, eval_inter, save_inter, model, data_train, data_val, data_test, optimizer, criterion_rt,
criterion_intensity, metric_rt, metric_intensity, forward, wandb=None):
for e in range(1, epochs + 1):
train(model, data_train, e, optimizer, criterion_rt, criterion_intensity, metric_rt, metric_intensity, forward,
wandb=wandb)
if e % eval_inter == 0:
eval(model, data_val, e, criterion_rt, criterion_intensity, metric_rt, metric_intensity, forward,
wandb=wandb)
if e % save_inter == 0:
save(model, 'model_common_' + str(e) + '.pt')
def main(args):
if args.wandb is not None:
os.environ["WANDB_API_KEY"] = 'b4a27ac6b6145e1a5d0ee7f9e2e8c20bd101dccd'
os.environ["WANDB_MODE"] = "offline"
os.environ["WANDB_DIR"] = os.path.abspath("./wandb_run")
wdb.init(project="Common prediction", dir='./wandb_run', name=args.wandb)
print(args)
print('Cuda : ', torch.cuda.is_available())
if args.forward == 'both':
data_train, data_val, data_test = common_dataset.load_data(path_train=args.dataset_train,
path_val=args.dataset_val,
path_test=args.dataset_test,
batch_size=args.batch_size, length=25, pad = False, convert=True, vocab='iapuc')
elif args.forward == 'rt':
data_train, data_val, data_test = dataloader.load_data(data_source=args.dataset_train,
batch_size=args.batch_size, length=25)
print('\nData loaded')
model = Model_Common_Transformer_TAPE(encoder_ff=args.encoder_ff, decoder_rt_ff=args.decoder_rt_ff,
decoder_int_ff=args.decoder_int_ff
, n_head=args.n_head, encoder_num_layer=args.encoder_num_layer,
decoder_int_num_layer=args.decoder_int_num_layer,
decoder_rt_num_layer=args.decoder_rt_num_layer, drop_rate=args.drop_rate,
embedding_dim=args.embedding_dim, acti=args.activation, norm=args.norm_first)
if torch.cuda.is_available():
model = model.cuda()
optimizer = optim.Adam(model.parameters(), lr=args.lr)
print('\nModel initialised')
run(epochs=args.epochs, eval_inter=args.eval_inter, save_inter=args.save_inter, model=model, data_train=data_train,
data_val=data_val, data_test=data_test, optimizer=optimizer, criterion_rt=torch.nn.MSELoss(),
criterion_intensity=masked_cos_sim, metric_rt=distance, metric_intensity=masked_spectral_angle,
wandb=args.wandb, forward=args.forward)
if args.wandb is not None:
wdb.finish()
def save(model, checkpoint_name):
print('\nModel Saving...')
os.makedirs('checkpoints', exist_ok=True)
torch.save(model, os.path.join('checkpoints', checkpoint_name))
def load(path):
model = torch.load(os.path.join('checkpoints', path))
return model
def get_n_params(model):
pp = 0
for n, p in list(model.named_parameters()):
nn = 1
for s in list(p.size()):
nn = nn * s
print(n, nn)
pp += nn
return pp
if __name__ == "__main__":
args = load_args()
main(args)