-
Notifications
You must be signed in to change notification settings - Fork 15
/
Copy pathtrainer.py
182 lines (135 loc) · 5.46 KB
/
trainer.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
import glob, os, math, time
import numpy as np
np.random.seed(0)
import tensorflow as tf
from tensorflow.keras import layers
from tensorflow.keras.models import Model
from sklearn.model_selection import KFold
from sklearn.metrics import accuracy_score
from sklearn.linear_model import LogisticRegression
from sklearn.decomposition import PCA
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import mean_squared_error
from pickle import dump
import models
def train_n_folds(model_type, data, config):
train_accuracies = []
val_accuracies = []
fold = 0
x = data[model_type]
if config.task == 'classification':
y = data['y_clf']
elif config.task == 'regression':
y = data['y_reg']
subjects = data['subjects']
if config.split_reference == 'samples':
splitter = x
elif config.split_reference == 'subjects':
splitter = subjects
if config.dataset_split == 'kfold':
for train_index, val_index in KFold(config.n_folds).split(splitter):
fold+=1
if config.split_reference == 'samples':
x_train, x_val = x[train_index], x[val_index]
y_train, y_val = y[train_index], y[val_index]
results = train_a_fold(model_type, x_train, y_train, x_val, y_val, fold, config)
train_accuracy, val_accuracy= results
train_accuracies.append(train_accuracy)
val_accuracies.append(val_accuracy)
else:
numpy_seeds = [913293, 653261, 84754, 645, 13451235]
for i in range(config.n_folds):
fold+=1
np.random.seed(numpy_seeds[i])
p = np.random.permutation(len(x))
x = x[p]
y = y[p]
n_train = int(config.split_ratio * len(x))
x_train, x_val = x[:n_train], x[n_train:]
y_train, y_val = y[:n_train], y[n_train:]
results = train_a_fold(model_type, x_train, y_train, x_val, y_val, fold, config)
train_accuracy, val_accuracy= results
train_accuracies.append(train_accuracy)
val_accuracies.append(val_accuracy)
return train_accuracies, val_accuracies
def train_a_fold(model_type, x_train, y_train, x_val, y_val, fold, config, sample_weight=None):
print('Training fold {} of {}'.format(fold, model_type))
if model_type == 'pause':
model = models.create_pause_model(config.task, config.n_pause_features, config.uncertainty)
epsilon = 1e-07
config.lr = 0.00125
config.batch_size = 24
elif model_type == 'intervention':
model = models.create_intervention_model(config.task, config.longest_speaker_length, config.uncertainty)
epsilon = 1e-07
config.lr = 0.00125
config.batch_size = 24
elif model_type == 'compare':
model = models.create_compare_model(config.task, config.compare_features_size, config.uncertainty)
epsilon = 1e-07
config.lr = 0.01
config.batch_size = 16
sc = StandardScaler()
sc.fit(x_train)
x_train = sc.transform(x_train)
x_val = sc.transform(x_val)
pca = PCA(n_components=config.compare_features_size)
pca.fit(x_train)
x_train = pca.transform(x_train)
x_val = pca.transform(x_val)
dump(sc, open(os.path.join(config.model_dir, 'compare/scaler_{}.pkl'.format(fold)), 'wb'))
dump(pca, open(os.path.join(config.model_dir, 'compare/pca_{}.pkl'.format(fold)), 'wb'))
elif model_type == 'silences':
model = models.create_silences_model(config.task, config.uncertainty)
epsilon = 1e-07
save_weights_only = False
if config.task == 'classification':
best_model = 'val_loss'
model.compile(loss= tf.keras.losses.categorical_crossentropy,
optimizer=tf.keras.optimizers.Adam(lr=config.lr, epsilon=epsilon),
metrics=['categorical_accuracy'])
elif config.task == 'regression':
best_model = 'val_loss'
if config.uncertainty:
def negloglik(y, p_y):
return -p_y.log_prob(y)
model.compile(loss=negloglik,
optimizer=tf.keras.optimizers.Adam(lr=config.lr, epsilon=epsilon), metrics=['mse'])
save_weights_only = False
else:
model.compile(loss=tf.keras.losses.mean_squared_error,
optimizer=tf.keras.optimizers.Adam(lr=config.lr, epsilon=epsilon), metrics=['mse'])
checkpointer = tf.keras.callbacks.ModelCheckpoint(
os.path.join(config.model_dir, model_type, 'fold_{}.h5'.format(fold)), monitor=best_model, verbose=0, save_best_only=True,
save_weights_only=save_weights_only, mode='auto', save_freq='epoch')
hist = model.fit(x_train, y_train,
batch_size=config.batch_size,
epochs=config.n_epochs,
verbose=config.verbose,
callbacks=[checkpointer],
validation_data=(x_val, y_val),
sample_weight=sample_weight)
if config.uncertainty:
model = tf.keras.models.load_model(os.path.join(config.model_dir, model_type, 'fold_{}.h5'.format(fold)),
custom_objects={'negloglik': negloglik})
preds = model(x_train)
mus = preds.mean().numpy()
train_score = mean_squared_error(y_train,mus, squared=False)
preds = model(x_val)
mus = preds.mean().numpy()
val_score = mean_squared_error(y_val, mus, squared=False)
else:
model = tf.keras.models.load_model(os.path.join(config.model_dir, model_type, 'fold_{}.h5'.format(fold)))
train_score = model.evaluate(x_train, y_train, verbose=0)
if config.task == 'classification':
train_score = train_score[1]
val_score = model.evaluate(x_val, y_val, verbose=0)
if config.task == 'classification':
val_score = val_score[1]
epoch_val_losses = hist.history['val_loss']
best_epoch_val_loss, best_epoch = np.min(epoch_val_losses), np.argmin(epoch_val_losses)+1
best_epoch_train_loss = hist.history['loss'][best_epoch-1]
print('Best Epoch: {:d}'.format(best_epoch))
print('Best Val loss {:.3f}'.format(best_epoch_val_loss))
print('Fold Val accuracy:', val_score)
return train_score, val_score