-
Notifications
You must be signed in to change notification settings - Fork 1
/
single_level_models.py
172 lines (150 loc) · 7.65 KB
/
single_level_models.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 heapq
import pandas as pd
import numpy as np
import os
import pickle
from sklearn import metrics
import matplotlib.pyplot as plt
from tensorflow import keras
import tensorflow as tf
import argparse
from sklearn.model_selection import train_test_split
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping
from tensorflow.keras.models import load_model
import neptune
from neptunecontrib.monitoring import utils
from neptunecontrib.monitoring.keras import NeptuneMonitor
from utils.util_models import getEpochNum, PredictiveModel, InnerPretrain
from utils.util_single_level import singleLevelModel, dataGeneratorSinLevelVariantHorizon
from utils.util_setup import *
from utils.util_evaluation import evaluate
from informer.informer_predictor import InformerPredictor
def train_outer_model(train_generator, test_generator, args):
'''Fit model'''
model_family = singleLevelModel(
max_n_actions=args.max_n_actions,
embedding_matrix=embedding_matrix,
dropout_rate=0.5
)
if args.model_name == 'simpletcn':
model = model_family.simpletcn()
elif args.model_name == 'restcn':
model = model_family.restcn()
elif args.model_name == 'gru':
model = model_family.gru()
elif args.model_name == 'fcn':
model = model_family.fcn()
else:
ValueError("Please specify model name!")
print(args)
# define the checkpoints
monitor = 'val_auc'
checkpoint = ModelCheckpoint(args.outer_save_path, monitor=monitor, verbose=1, save_best_only=True, mode='max')
earlystopping = EarlyStopping(monitor=monitor, mode='max', verbose=1, patience=20)
callback_list = [checkpoint]
if args.if_earlystopping: callback_list.append(earlystopping)
if args.if_neptune: callback_list += [NeptuneMonitor(), getEpochNum()]
# train model
model.fit_generator(generator=train_generator,
validation_data=test_generator,
epochs=args.epochs,
use_multiprocessing=True, #int(args.gpu_id) == -1, # multiple CPUs for data preprocessing
workers=args.workers, # multiple CPUs for calculating
callbacks=callback_list,
verbose=1)
def evaluateVariantHorizon(best_model, args, df_survey, IDs_test, embedding_matrix, horizon=0):
predictor = keras.models.Model(inputs=best_model.input, outputs=best_model.get_layer('prediction').output)
predictor.compile(
loss='categorical_crossentropy',
metrics=['accuracy', tf.keras.metrics.AUC(curve='ROC'), tf.keras.metrics.AUC(curve='PR')]
)
print('-----------Test horizon {} -----------'.format(horizon))
test_generator = dataGeneratorSinLevelVariantHorizon(df_survey, IDs_test,
batch_size=1,
dim=args.max_n_actions,
vocab_size=embedding_matrix.shape[0],
shuffle=False,
if_deepsup=False,
horizon=horizon)
# Get testing scores
print('Testing model...')
y_test = []
for i in range(len(test_generator)):
y_i = test_generator[i][1][:, 1]
y_test += list(y_i)
y_test = np.array(y_test)
y_prob = predictor.predict_generator(test_generator, steps=len(test_generator), verbose=1)[:, 1]
# print(y_prob)
# evaluation
fpr, tpr, _ = metrics.roc_curve(y_test, y_prob)
prec, rec, _ = metrics.precision_recall_curve(y_test, y_prob)
acc = metrics.accuracy_score(y_test, y_prob > 0.5)
print('--------------------------------------------')
print('Evaluation of full test set:')
print("AU-ROC:", "%0.4f" % metrics.auc(fpr, tpr),
"AU-PRC:", "%0.4f" % metrics.auc(rec, prec),
"Accuracy:", "%0.4f" % acc)
if args.if_neptune:
neptune.log_metric('AU-ROC', metrics.auc(fpr, tpr))
neptune.log_metric('AU-PRC', metrics.auc(rec, prec))
neptune.log_metric('Accuracy', acc)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--memo', type=str, default='CV_single_resTCN_woActEmb')
parser.add_argument('--max_n_actions', type=int, default=50000)
parser.add_argument('--dim_embedding', type=int, default=100)
parser.add_argument('--epochs', type=int, default=10)
parser.add_argument('--batch_size', type=int, default=1)
parser.add_argument('--workers', type=int, default=5)
parser.add_argument('--tcn_layers', type=int, default=6)
parser.add_argument('--lr', type=float, default=0.001)
parser.add_argument('--gpu_id', type=str, default='0')
parser.add_argument('--if_pretrain', type=boolean_string, default=False)
parser.add_argument('--if_deepsup', type=boolean_string, default=False)
parser.add_argument('--if_train_predictor', type=boolean_string, default=False)
parser.add_argument('--if_neptune', type=boolean_string, default=False)
parser.add_argument('--if_earlystopping', type=boolean_string, default=True)
parser.add_argument('--if_mem_constr', type=boolean_string, default=True)
parser.add_argument('--if_flatten', type=boolean_string, default=True)
parser.add_argument('--if_static', type=boolean_string, default=False)
parser.add_argument('--trial_idx', type=int, default=5)
parser.add_argument('--cv_idx', type=int, default=0)
parser.add_argument('--model_name', type=str, default='tcn')
parser.add_argument('--horizon', type=int, default=5)
args = parser.parse_args()
print(args)
# Use GPU for training
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu_id
if args.if_mem_constr:
physical_devices = tf.config.list_physical_devices('GPU')
try: tf.config.experimental.set_memory_growth(physical_devices[0], True)
except: pass
# log experiment with Neptune and set random seed
if args.if_neptune: init_neptune(args)
set_random_seeds(123)
# load survey and embeddings, make dir
survey = pd.read_csv('data_processed/survey_processed.csv')
survey = survey[survey['burnout_bin'].notna()]
embedding_matrix = get_embeddings(args.dim_embedding)
# model path
model_folder = os.path.join('model', args.memo)
path_suffix = '_trial_' + str(args.trial_idx) + '_fold_' + str(args.cv_idx)
if not os.path.exists(model_folder): os.mkdir(model_folder)
args.outer_save_path = os.path.join(model_folder, "higher" + path_suffix)
# train-validation split
print('\n--------- Trial index: {} ---------'.format(args.trial_idx))
print('\nFold index (0-4): {}'.format(args.cv_idx))
cv = RandomizedGroupKFold(groups=survey['USER_ID'].to_numpy(), n_splits=5, random_state=args.trial_idx)
train_ix, test_ix = cv[args.cv_idx]
train_uids = survey.iloc[train_ix, :]['USER_ID'].unique().tolist()
test_uids = survey.iloc[test_ix, :]['USER_ID'].unique().tolist()
# prepare data for higher model
_, df_survey, IDs = data_loader_outer(train_uids, test_uids)
print('------------------\nIncluded surveys:', df_survey.sample_ID.nunique(), '\n-------------------')
train_generator, valid_generator, test_generator = get_generator_single(args, embedding_matrix, IDs, df_survey)
# train higher model
if args.if_train_predictor: train_outer_model(train_generator, valid_generator, args)
best_model = load_model(args.outer_save_path)
evaluate(best_model, test_generator, args)
# evaluateVariantHorizon(best_model, args, df_survey, IDs[-1], embedding_matrix, horizon=args.horizon)