-
Notifications
You must be signed in to change notification settings - Fork 2
/
finetune_all.py
336 lines (284 loc) · 14.2 KB
/
finetune_all.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
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
import numpy as np
import pandas as pd
import torch
import csv
import random
from sklearn.model_selection import train_test_split
import numpy as np
from sklearn.metrics import log_loss, roc_auc_score
from sklearn.preprocessing import LabelEncoder, MinMaxScaler
from tensorflow.keras.preprocessing.sequence import pad_sequences
from deepctr_torch.callbacks import EarlyStopping, ModelCheckpoint
from preprocessing.inputs import SparseFeat, get_feature_names
from dataset import CtrDataset2
from transformers import AutoModel, AutoTokenizer
from utils import AvgMeter, get_lr, check_path
from tqdm import tqdm
from torch import nn
from layers.core import concat_fun
from finetune_config import create_finetune_all_parser
from preprocessing.inputs import combined_dnn_input
from layers.core import PredictionLayer
# finetune all
class DCNMix_NLP_Model(nn.Module):
def __init__(self):
super(DCNMix_NLP_Model, self).__init__()
self.model = torch.load(load_pretrain_path)
self.text_dense = nn.Linear(cfg.text_embedding_dim,1)
self.text_out = PredictionLayer()
self.out = PredictionLayer()
# self.alpha = torch.tensor(cfg.alpha)
self.alpha = torch.nn.Parameter(torch.zeros((1,)), requires_grad=True)
def forward(self, batch):
sparse_embedding_list, dense_value_list = self.model.rec_encoder.input_from_feature_columns(batch["rec_data"], self.model.rec_encoder.dnn_feature_columns,
self.model.rec_encoder.embedding_dict)
logit = self.model.rec_encoder.linear_model(batch["rec_data"])
dnn_input = combined_dnn_input(sparse_embedding_list, dense_value_list)
deep_out = self.model.rec_encoder.dnn(dnn_input)
cross_out = self.model.rec_encoder.crossnet(dnn_input)
stack_out = torch.cat((cross_out, deep_out), dim=-1)
logit += self.model.rec_encoder.dnn_linear(stack_out)
text_features = self.model.text_encoder(
input_ids=batch["input_ids"], attention_mask=batch["attention_mask"]
)
text_logit = self.text_dense(text_features)
if cfg.mixed_precision:
ctr_pred = logit
text_pred = text_logit
y_pred = logit*self.alpha + text_logit*(1-self.alpha)
else:
ctr_pred = self.model.rec_encoder.out(logit)
text_pred = self.text_out(text_logit)
y_pred = self.out(logit*self.alpha + text_logit*(1-self.alpha))
return ctr_pred, text_pred, y_pred
def add_regularization_weight(regularization_weight, weight_list, l1=0.0, l2=0.0):
if isinstance(weight_list, torch.nn.parameter.Parameter):
weight_list = [weight_list]
else:
weight_list = list(weight_list)
regularization_weight.append((weight_list, l1, l2))
return regularization_weight
def get_regularization_loss(regularization_weight):
total_reg_loss = torch.zeros((1,), device='cuda')
for weight_list, l1, l2 in regularization_weight:
for w in weight_list:
if isinstance(w, tuple):
parameter = w[1] # named_parameters
else:
parameter = w
if l2 > 0:
try:
total_reg_loss += torch.sum(l2 * torch.square(parameter))
except AttributeError:
total_reg_loss += torch.sum(l2 * parameter * parameter)
return total_reg_loss
def make_train_valid_dfs(struct_data_path, text_data_path, seed, data_source):
struct_data = pd.read_csv(struct_data_path)
text_data = pd.read_table(text_data_path,names = ["content"],header=None)
if cfg.sample:
struct_data,_ = train_test_split(struct_data,test_size= (1-cfg.sample_ration) ,random_state= seed)
text_data,_ = train_test_split(text_data,test_size= (1-cfg.sample_ration) ,random_state= seed)
text_data['label'] = struct_data['label']
train_struct, test_struct = struct_data.iloc[:int(len(struct_data) * 0.9)].copy(), struct_data.iloc[int(len(
struct_data) * 0.9):].copy()
train_text, test_text = text_data.iloc[:int(len(text_data) * 0.9)].copy(), text_data.iloc[int(len(
text_data) * 0.9):].copy()
print('train size, test size: ', len(train_struct), len(test_struct))
return train_struct, test_struct, train_text, test_text, struct_data
def build_loaders(struct_input, text_input, linear_feature_columns,dnn_feature_columns,tokenizer, mode):
dataset = CtrDataset2(
struct_input,
text_input["content"].values,
text_input["label"].values,
linear_feature_columns=linear_feature_columns,
dnn_feature_columns=dnn_feature_columns,
tokenizer=tokenizer,
max_length=cfg.max_length
)
dataloader = torch.utils.data.DataLoader(
dataset,
batch_size=cfg.batch_size,
num_workers=cfg.num_workers,
pin_memory = True,
shuffle=True if mode == "train" else False,
)
return dataloader
def process_struct_data(data_source, train, test, data):
embedding_dim = 32
if (data_source == 'movielens'):
sparse_features = ['user_id', 'gender', 'age', 'occupation', 'zipcode', 'movie_id', 'title','genre']
elif (data_source == 'bookcrossing'):
sparse_features = ['User ID', 'Location', 'Age', 'ISBN', 'Book title', 'Author', 'Publication year', 'Publisher']
elif (data_source == 'goodreads'):
sparse_features = ['User ID','Book ID', 'Book title', 'Book genres' ,'Average rating', 'Number of book reviews', 'Author ID', 'Author name',
'Number of pages','eBook flag', 'Format', 'Publisher', 'Publication year', 'Work ID', 'Media type']
sparse_feature_columns = [SparseFeat(feat, data[feat].nunique(), embedding_dim=embedding_dim)
for i, feat in enumerate(sparse_features)]
linear_feature_columns = sparse_feature_columns
dnn_feature_columns = sparse_feature_columns
train_model_input = {name: train[name] for name in sparse_features }
test_model_input = {name: test[name] for name in sparse_features }
return linear_feature_columns, dnn_feature_columns, train_model_input, test_model_input
def train_epoch(model, train_loader, optimizer, step,loss_fnc):
loss_meter = AvgMeter()
tqdm_object = tqdm(train_loader, total=len(train_loader))
scaler = GradScaler()
predicts = []
labels = []
for batch in tqdm_object:
batch = {k: v.to(cfg.device) for k, v in batch.items() if k != "text_data"}
label = batch['label'].unsqueeze(1).to(cfg.device)
if cfg.mixed_precision:
optimizer.zero_grad()
with autocast():
ctr_output, text_output, output = model(batch)
reg_loss = get_regularization_loss(regularization_weight)
loss = loss_fnc(output, label.float()) + loss_fnc(ctr_output, label.float()) + loss_fnc(text_output, label.float())+ reg_loss
scaler.scale(loss).backward()
scaler.step(optimizer)
scaler.update()
predicts.extend(torch.sigmoid(output).cpu().data.numpy())
labels.extend(label.cpu().data.numpy())
else:
optimizer.zero_grad()
ctr_output, text_output, output = model(batch)
reg_loss = get_regularization_loss(regularization_weight)
loss = loss_fnc(output, label.float()) + loss_fnc(ctr_output, label.float()) + loss_fnc(text_output, label.float()) + reg_loss
loss.backward()
optimizer.step()
predicts.extend(output.cpu().data.numpy())
labels.extend(label.cpu().data.numpy())
count = batch["label"].size(0)
loss_meter.update(loss.item(), count)
auc = roc_auc_score(label.cpu().data.numpy(), output.cpu().data.numpy())
tqdm_object.set_postfix(train_loss=loss_meter.avg, lr=get_lr(optimizer))
labels = np.concatenate(labels).astype(np.float64)
predicts = np.concatenate(predicts).astype(np.float64)
auc = roc_auc_score(labels, predicts)
m2 = log_loss(labels, predicts)
print(f"train auc:{auc}, train logloss:{m2}")
return loss_meter, auc, m2
def valid_epoch(model, valid_loader,loss_fnc):
loss_meter = AvgMeter()
tqdm_object = tqdm(valid_loader, total=len(valid_loader))
predicts = []
labels = []
for batch in tqdm_object:
batch = {k: v.to(cfg.device) for k, v in batch.items() if k != "text_data"}
label = batch['label'].unsqueeze(1).to(cfg.device)
ctr_output, text_output, output = model(batch)
loss = loss_fnc(output, label.float()) + loss_fnc(ctr_output, label.float()) + loss_fnc(text_output, label.float())
count = batch["label"].size(0)
if cfg.mixed_precision:
predicts.extend(torch.sigmoid(output).cpu().data.numpy())
else:
predicts.extend(output.cpu().data.numpy())
labels.extend(label.cpu().data.numpy())
loss_meter.update(loss.item(), count)
tqdm_object.set_postfix(valid_loss=loss_meter.avg)
labels = np.concatenate(labels).astype(np.float64)
predicts = np.concatenate(predicts).astype(np.float64)
auc = roc_auc_score(labels, predicts)
m2 = log_loss(labels, predicts)
print(f"valid auc:{auc}, valid logloss:{m2}")
return loss_meter, auc, m2
def setup_seed(seed):
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
random.seed(seed)
torch.backends.cudnn.deterministic = True
if __name__ == '__main__':
seed = 2012
cfg = create_finetune_all_parser()
data_source = cfg.dataset
model = None
setup_seed(seed)
train_struct, test_struct, train_text, test_text, struct_data = make_train_valid_dfs(cfg.struct_path,cfg.text_path,seed, data_source)
linear_feature_columns, dnn_feature_columns, train_struct_input, test_struct_input = \
process_struct_data(data_source,train_struct,test_struct, struct_data)
tokenizer = AutoTokenizer.from_pretrained(cfg.text_tokenizer,local_files_only=True)
train_loader = build_loaders(train_struct_input, train_text,
linear_feature_columns,dnn_feature_columns,tokenizer, mode='train')
test_loader = build_loaders(test_struct_input, test_text,
linear_feature_columns,dnn_feature_columns,tokenizer, mode ='test')
load_pretrain_path = cfg.load_prefix_path + \
f'Feature_restore_model/{cfg.model_path}/{data_source}{cfg.temperature}_{cfg.use_mfm}_{cfg.use_mlm}_{cfg.pre_epochs}_{cfg.pre_lr}_0.15_0.15_best.pt'
print('load pretrain path', load_pretrain_path)
save_path = f'Feature_finetune_all_models/{cfg.model_path}/'
write_path = f'Feature_finetune_all_results/{cfg.model_path}/'
check_path(save_path)
check_path(write_path)
save_path += f'{data_source}_{cfg.temperature}_{cfg.use_mfm}_{cfg.use_mlm}_{cfg.pre_epochs}_{cfg.pre_lr}_0.15_0.15.pt'
write_path += f'{data_source}_{cfg.temperature}_{cfg.use_mfm}_{cfg.use_mlm}_{cfg.pre_epochs}_{cfg.pre_lr}_0.15_0.15.txt'
print('begin finetune all')
if cfg.dataset == 'movielens':
l2_reg = 1e-4
elif cfg.dataset == 'bookcrossing':
l2_reg = 1e-4
elif cfg.dataset == 'goodreads':
l2_reg = 0
model_name = cfg.backbone
if model_name == 'DCNv2':
model = DCNMix_NLP_Model()
model.to(cfg.device)
model = torch.nn.DataParallel(model, device_ids=range(torch.cuda.device_count()))
text_param_list, other_param_list = [],[]
for (name, param) in model.module.named_parameters():
if 'text_encoder' in name:
text_param_list.append(param)
else:
other_param_list.append(param)
params = [
{"params": text_param_list, "lr": 5e-5},
{"params": other_param_list}
]
optimizer = torch.optim.AdamW(
params, lr=cfg.lr, weight_decay=cfg.weight_decay
)
print('l2 reg', l2_reg)
regularization_weight = []
add_regularization_weight(regularization_weight, model.module.model.rec_encoder.embedding_dict.parameters(), l2=l2_reg)
add_regularization_weight(regularization_weight, model.module.model.rec_encoder.linear_model.parameters(), l2=l2_reg)
# print(len(regularization_weight))
if cfg.mixed_precision:
loss_fn = nn.BCEWithLogitsLoss()
else:
loss_fn = nn.BCELoss()
best_auc = -float('inf')
step = "epoch"
print("begain Training")
patience = 3
es_cnt = 0
early_stop = False
for epoch in range(cfg.epochs):
if early_stop:
print('early stop')
break
print(f"Epoch: {epoch + 1}")
print(optimizer.state_dict()['param_groups'][0]['lr'])
model.train()
train_loss, train_auc, train_logloss = train_epoch(model,
train_loader,
optimizer,
step, loss_fn)
# lr_scheduler.step()
model.eval()
with torch.no_grad():
valid_loss, valid_auc, valid_logloss = valid_epoch(model, test_loader, loss_fn)
if valid_auc > best_auc:
best_auc = valid_auc
es_cnt = 0
torch.save(model.module, save_path)
print('save model')
else:
es_cnt += 1
if es_cnt >= patience:
early_stop = True
print('train auc {:.5f}, train logloss {:.5f}, valid auc {:.5f}, valid logloss {:.5f}, best auc {:.5f}'.format(
train_auc, train_logloss, valid_auc, valid_logloss, best_auc))
tmp_alpha = model.module.alpha.cpu().detach().numpy().item()
print("alpha {:.5f}".format(tmp_alpha))
writer_text = [epoch, l2_reg, cfg.lr,cfg.batch_size,train_auc, train_logloss, valid_auc, valid_logloss, best_auc, tmp_alpha]
with open(write_path,'a+') as writer:
writer.write(' '.join([str(x) for x in writer_text]) + '\n')