-
Notifications
You must be signed in to change notification settings - Fork 16
/
Copy pathmodel_manager.py
191 lines (167 loc) · 9.68 KB
/
model_manager.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
import os
from tqdm import tqdm
import sys
import numpy as np
import preprocessing
import tensorflow as tf
import model
import data_loader
import pickle
import matplotlib.pyplot as plt
from eval import *
class ModelManager():
def __init__(self, model_name, input_length, num_class, dim, lr, seed, K=5, is_LLAL=False):
self.model_name = model_name
self.input_length = input_length
self.num_class = num_class
self.dim = dim
self.seed = seed
self.K = K
self.model_init = 0
self.is_LLAL = is_LLAL
if self.model_name == "TCN":
self.model = model.TCN(self.num_class, self.dim, lr)
elif self.model_name == "LongTCN":
self.model = model.TCN(self.num_class, self.dim, lr=lr, num_block=1, num_dilation=15)
elif self.model_name == "RNN":
self.model = model.RNN(self.num_class, self.dim, lr=lr)
elif self.model_name == "MSTCN":
self.model = model.MSTCN(self.num_class, self.dim, lr=lr, is_LLAL=is_LLAL)
elif self.model_name == "SSTCN":
self.model = model.MSTCN(self.num_class, self.dim, lr=lr, num_stage=1, num_dilation=5, is_LLAL=is_LLAL)
else:
print("wrong model name in ModelManager")
if is_LLAL:
self.model.input_LLAL(np.zeros((1, self.input_length, self.dim)))
self.model.call_LLAL(np.zeros((1, self.input_length, self.model.num_stage*self.model.num_filters)))
else:
self.model(np.zeros((1, self.input_length, self.dim)))
self.weights_init = self.model.get_weights()
def train_test_generator(self, X, y, y_seg, mask, file_boundaries):
assert(self.seed<=self.K-1)
self.test_data_start = len(X) // self.K * self.seed
if self.seed == self.K-1:
self.test_data_end = len(X)
else:
self.test_data_end = len(X) // self.K * (self.seed + 1)
self.X_long_train = np.concatenate([X[:self.test_data_start],X[self.test_data_end:]])
self.y_long_train = np.concatenate([y[:self.test_data_start],y[self.test_data_end:]])
self.y_seg_long_train = np.concatenate([y_seg[:self.test_data_start],y_seg[self.test_data_end:]])
self.mask_long_train = np.concatenate([mask[:self.test_data_start],mask[self.test_data_end:]])
self.file_boundaries_train = np.concatenate([file_boundaries[:self.test_data_start],file_boundaries[self.test_data_end:]])
self.X_long_test = X[self.test_data_start:self.test_data_end]
self.y_long_test = y[self.test_data_start:self.test_data_end]
self.y_seg_long_test = y_seg[self.test_data_start:self.test_data_end]
self.mask_long_test = mask[self.test_data_start:self.test_data_end] # TODO: labeled_or_not index check needed
self.file_boundaries_test = file_boundaries[self.test_data_start:self.test_data_end]
print(self.X_long_train.shape, self.y_long_train.shape, self.y_seg_long_train.shape, self.mask_long_train.shape,
self.file_boundaries_train.shape)
return self.X_long_train, self.y_long_train, self.y_seg_long_train, self.mask_long_train, self.file_boundaries_train, self.X_long_test, self.y_long_test, self.y_seg_long_test, self.mask_long_test, self.file_boundaries_test
def load_train_data(self, X, y, y_seg, mask, file_boundaries, fully_supervised=False):
self.dataloader = data_loader.DataLoader(self. input_length, X, y, y_seg, mask, file_boundaries, self.seed, fully_supervised)
def get_unlabeled_ECE(self):
output_final = self.model.predict(X_long=self.X_long_train,file_boundaries=self.file_boundaries_train)
ECE = 0
pred = np.argmax(output_final, axis=1)
conf = np.max(output_final, axis=1)
prev_i = 0
for i in [0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]:
if np.sum((conf>prev_i) & (conf<i)) == 0:
prev_i = i
continue
conf_bin = np.mean(conf[(conf>prev_i) & (conf<i)])
acc_bin = np.sum(pred[(conf>prev_i) & (conf<i)]==self.y_long_train[(conf>prev_i) & (conf<i)])/np.sum((conf>prev_i) & (conf<i))
ECE += np.sum((conf>prev_i) & (conf<i))/len(conf)*np.abs(acc_bin-conf_bin)
prev_i = i
return ECE
def test_model(self, bg_class = []):
file_boundary_ind = np.where(self.file_boundaries_test==1)[0].tolist()
start = 0 # test_data_start_ind
if len(file_boundary_ind) > 0:
if not len(self.file_boundaries_test)-1 in file_boundary_ind:
file_boundary_ind.append(len(self.file_boundaries_test)-1)
for i in file_boundary_ind:
output_final_file = self.model(self.X_long_test[np.newaxis, start:i+1]).numpy()[0]
if start == 0:
output_final = output_final_file
else:
output_final = np.concatenate([output_final, output_final_file], axis=0)
start = i+1
else:
output_final = self.model(self.X_long_test[np.newaxis, :, :]).numpy()
output_final = output_final.reshape((-1, self.num_class))
y_test_flatten = tf.reshape(self.y_long_test, [-1]).numpy()
if len(output_final)!=len(y_test_flatten):
print(self.y_long_test.shape, self.X_long_test.shape, len(self.file_boundaries_test), file_boundary_ind, len(output_final), len(y_test_flatten))
print("shapes are different when testing")
return get_all_metrics(np.argmax(output_final,axis=1),y_test_flatten,bg_class=bg_class)
def train_model_new(self, epoch, batch_size, is_test=False):
self.model.set_weights(self.weights_init)
prob, indice_list = self.dataloader.window_scoring(slide_size=self.input_length)
num_steps_per_epoch = len(prob) // batch_size # epoch is set by # of proped+queried labels
train_loss = []
test_acc=[]
# for i in tqdm(range(epoch), leave=True, desc="train_model"):
least_train_loss = np.inf
for i in tqdm(range(epoch), leave=True, desc="train_model"):
train_loss_per_epoch = 0
idxs_prob = np.random.choice(len(prob), batch_size*num_steps_per_epoch, p=prob)
idxs = indice_list[idxs_prob].tolist()
dataset = self.dataloader.dataset_generator(idxs, batch_size)
for X_windowed, y_windowed, y_seg_windowed, mask_windowed in dataset: # should be substituted by [for x,y,mask in dataset: ...]
if self.is_LLAL and i<=37:
loss_final = self.model.train_step_llal_full(X_windowed, y_windowed, mask_windowed)
elif self.is_LLAL and i>37:
loss_final = self.model.train_step_llal_part(X_windowed, y_windowed, mask_windowed)
else:
loss_final = self.model.train_step(X_windowed, y_windowed, mask_windowed)
train_loss_per_epoch += loss_final.numpy()
if is_test:
metric = self.test_model()
print(i, train_loss_per_epoch, metric)
test_acc.append(metric)
if least_train_loss > train_loss_per_epoch:
weights_least_loss = self.model.get_weights()
least_train_loss = train_loss_per_epoch
train_loss.append(train_loss_per_epoch)
self.model.set_weights(weights_least_loss)
return train_loss, test_acc
def train_model(self, epoch, batch_size, is_test=False):
self.model.set_weights(self.weights_init)
prob, indice_list = self.dataloader.window_scoring(slide_size=self.input_length)
num_steps_per_epoch = len(prob) // batch_size # epoch is set by # of proped+queried labels
train_loss = []
test_acc=[]
# for i in tqdm(range(epoch), leave=True, desc="train_model"):
least_train_loss = np.inf
for i in tqdm(range(epoch), leave=True, desc="train_model"):
train_loss_per_epoch = 0
for j in range(num_steps_per_epoch):
indice = np.random.choice(len(prob), batch_size, p=prob) # oversampling
X_windowed, y_windowed, y_seg_windowed, mask_windowed = self.dataloader.batch_generator(indice_list[indice].tolist())
if self.is_LLAL and i<=37:
loss_final = self.model.train_step_llal_full(X_windowed, y_windowed, mask_windowed)
elif self.is_LLAL and i>37:
loss_final = self.model.train_step_llal_part(X_windowed, y_windowed, mask_windowed)
else:
loss_final = self.model.train_step(X_windowed, y_windowed, mask_windowed)
train_loss_per_epoch += loss_final.numpy()
if is_test:
metric = self.test_model()
print(i, train_loss_per_epoch, metric)
test_acc.append(metric)
if least_train_loss > train_loss_per_epoch:
weights_least_loss = self.model.get_weights()
least_train_loss = train_loss_per_epoch
train_loss.append(train_loss_per_epoch)
self.model.set_weights(weights_least_loss)
return train_loss, test_acc
if __name__ == "__main__":
# python3 model_manager.py gpu data_name model_name input_length learning_rate
with tf.device("/GPU:"+sys.argv[1]):
data = preprocessing.Preprocessing(sys.argv[2], 0.1)
X_long, y_long, y_seg_long, file_boundaries = data.generate_long_time_series()
MM = ModelManager(model_name=sys.argv[3], input_length=int(sys.argv[4]), num_class=len(np.unique(y_long)),
dim=X_long.shape[1], lr=float(sys.argv[5]))
MM.load_train_data(X=X_long, y=y_long, y_seg=y_seg_long, mask=np.ones(len(y_long)), file_boundaries=file_boundaries)
train_loss, test_acc = MM.train_model(1000, 32, y_long)