-
Notifications
You must be signed in to change notification settings - Fork 2
/
spam_random.py
273 lines (230 loc) · 10.3 KB
/
spam_random.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
import torch
import sys
import numpy as np
from logistic_regression import *
from sklearn.metrics import f1_score, accuracy_score
from cage import *
from sklearn.feature_extraction.text import TfidfVectorizer
from losses import *
import pickle
import apricot
import statistics
torch.set_default_tensor_type(torch.DoubleTensor)
def load_pickle(file_name):
with open(file_name, 'rb') as f:
objs = []
while 1:
try:
o = pickle.load(f)
except EOFError:
break
objs.append(o)
return objs
def find_indices(data, data_sub):
indices = []
for ele in data_sub:
x = np.where(data == ele)[0]
indices.append(statistics.mode(x))
return indices
if torch.cuda.is_available():
dev = "cuda:0"
else:
dev = "cpu"
device = torch.device(dev)
torch.set_printoptions(precision=20)
test_objs = load_pickle("Data/SMS/test_processed.p")
train_objs = load_pickle("Data/SMS/d_processed.p")
u_objs = load_pickle("Data/SMS/U_processed.p")
x_train = torch.from_numpy(train_objs[0]).double()
y_train = torch.from_numpy(train_objs[3])
x_test = torch.from_numpy(test_objs[0]).double()
y_test = torch.from_numpy(test_objs[3])
x_u = torch.from_numpy(u_objs[0]).double()
y_u = torch.from_numpy(u_objs[3])
n_classes = 2
n_lfs = 73
n_features = x_train.shape[1]
train_instances = x_train.shape[0]
test_instances = x_test.shape[0]
u_instances = x_u.shape[0]
# Discrete lambda values
l = torch.from_numpy(train_objs[2]).long()
u_l = torch.from_numpy(u_objs[2]).long()
l_test = torch.from_numpy(test_objs[2]).long()
# Continuous score values
s = torch.ones(train_instances, n_lfs).double()
s_u = torch.ones(u_instances, n_lfs).double()
s_test = torch.ones(test_instances, n_lfs).double()
# Labeling Function Classes
k = torch.from_numpy(np.array(
[1, 0, 1, 0, 0, 0, 0, 1, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 0, 1, 1, 1, 1, 0, 1, 0, 0, 0, 0, 0, 1,
0, 0, 0, 0, 0, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0])).long()
# True y
y_true = y_train
y_true_test = y_test
continuous_mask = torch.zeros(n_lfs).double()
a = torch.ones(n_lfs).double() * 0.9
for i in range(s.shape[0]):
for j in range(s.shape[1]):
if s[i, j].item() > 0.999:
s[i, j] = 0.999
if s[i, j].item() < 0.001:
s[i, j] = 0.001
for i in range(s_test.shape[0]):
for j in range(s_test.shape[1]):
if s_test[i, j].item() > 0.999:
s_test[i, j] = 0.999
if s_test[i, j].item() < 0.001:
s_test[i, j] = 0.001
n_supervised = int(len(y_true) * 1)
# n_supervised = int(1842)
l_unsupervised = u_l # [n_supervised:]
s_unsupervised = s_u # [n_supervised:]
features_unsupervised = x_u # x_train[n_supervised:]
y_unsupervised = torch.tensor(
y_u).long() # torch.tensor(y_torch.tensor(y_true[n_supervised:]).long()#true[n_supervised:]).long()
y_supervised = torch.tensor(y_true[:n_supervised]).long()
features_supervised = x_train[:n_supervised]
l_supervised = l[:n_supervised]
s_supervised = s[:n_supervised]
print('l_unsupervised', l_unsupervised.shape)
print('s_unsupervised', s_unsupervised.shape)
print('l_supervised', l_supervised.shape)
print('s_supervised', s_supervised.shape)
# total_count = y_unsupervised.size() + y_supervised.size()
X = torch.cat((features_supervised, features_unsupervised), 0)
y = torch.cat((y_supervised, y_unsupervised), 0)
#X_sub, y_sub = fl.fit_transform(X, y)
#indices = find_indices(X, X_sub)
lr_acc_list = []
lr_mac_list = []
gm_acc_list = []
gm_mac_list = []
for run in range(5):
pi = torch.ones((n_classes, n_lfs)).double()
pi.requires_grad = True
theta = torch.ones((n_classes, n_lfs)).double() * 1
theta.requires_grad = True
pi_y = torch.ones(n_classes).double()
pi_y.requires_grad = True
lr_model = LogisticRegression(n_features, n_classes)
optimizer = torch.optim.Adam([{"params": lr_model.parameters()}, {"params": [pi, pi_y, theta]}], lr=0.008)
# optimizer = torch.optim.Adam([theta, pi, pi_y], lr=0.01, weight_decay=0)
supervised_criterion = torch.nn.CrossEntropyLoss()
indices = np.random.choice(np.arange(len(X)), replace=False)
l = torch.cat((l_supervised, l_unsupervised), 0)
s = torch.cat((s_supervised, s_unsupervised), 0)
r = torch.zeros(X.shape[0])
r[indices] = 1
idx = np.arange(X.shape[0])
rand_idx = np.random.choice(idx, X.shape[0], replace=False)
X = X[rand_idx]
y = y[rand_idx]
l = l[rand_idx]
s = s[rand_idx]
r = r[rand_idx]
print('X', X.shape)
print('y', y.shape)
print('l', l.shape)
print('s', s.shape)
print('r', r.shape)
# print('s_supervised', s_supervised.shape)
# print('x_supervised', features_supervised.shape)
# print('y_supervised', y_supervised.shape)
gm_mac, gm_acc, lr_mac, lr_acc = 0, 0, 0, 0
BATCH_SIZE = 100
for epoch in range(100):
for i in range(int(np.floor(len(features_unsupervised) / BATCH_SIZE))):
if (i + 1) * BATCH_SIZE < len(features_unsupervised):
X_batch = X[(i * BATCH_SIZE):((i + 1) * BATCH_SIZE)]
l_batch = l[(i * BATCH_SIZE):((i + 1) * BATCH_SIZE)]
s_batch = s[(i * BATCH_SIZE):((i + 1) * BATCH_SIZE)]
y_batch = y[(i * BATCH_SIZE):((i + 1) * BATCH_SIZE)]
r_batch = r[(i * BATCH_SIZE):((i + 1) * BATCH_SIZE)]
# print('X_batch', X_batch.shape)
# print('y_batch', y_batch.shape)
# print('l_batch', l_batch.shape)
# print('s_batch', s_batch.shape)
# print('r', r_batch)
r_arr = r_batch.numpy()
X_sup_batch = X_batch[np.where(r_arr == 1)]
X_unsup_batch = X_batch[np.where(r_arr == 0)]
y_sup_batch = y_batch[np.where(r_arr == 1)]
y_unsup_batch = y_batch[np.where(r_arr == 0)]
l_sup_batch = l_batch[np.where(r_arr == 1)]
l_unsup_batch = l_batch[np.where(r_arr == 0)]
s_sup_batch = s_batch[np.where(r_arr == 1)]
s_unsup_batch = s_batch[np.where(r_arr == 0)]
# print('X_sup_batch.shape, y_sup_batch',X_sup_batch.shape[0], y_sup_batch.shape)
else:
X_batch = X[i * BATCH_SIZE:]
l_batch = y[i * BATCH_SIZE:]
s_batch = s[i * BATCH_SIZE:]
y_batch = y[i * BATCH_SIZE:]
r_batch = r[i * BATCH_SIZE:]
X_sup_batch = X_batch[np.where(r_arr == 1)]
X_unsup_batch = X_batch[np.where(r_arr == 0)]
y_sup_batch = y_batch[np.where(r_arr == 1)]
y_unsup_batch = y_batch[np.where(r_arr == 0)]
l_sup_batch = l_batch[np.where(r_arr == 1)]
l_unsup_batch = l_batch[np.where(r_arr == 0)]
s_sup_batch = s_batch[np.where(r_arr == 1)]
s_unsup_batch = s_batch[np.where(r_arr == 0)]
# print('X_sup_batch.shape, y_sup_batch',X_sup_batch.shape, y_sup_batch)
lr_model.train()
optimizer.zero_grad()
if X_sup_batch.shape[0] != 0:
loss_1 = supervised_criterion(lr_model(X_sup_batch), y_sup_batch)
loss_4 = log_likelihood_loss_supervised(theta, pi_y, pi, y_sup_batch, l_sup_batch, s_sup_batch, k,
n_classes, continuous_mask)
unsupervised_lr_probability = torch.nn.Softmax()(lr_model(X_unsup_batch))
loss_2 = entropy(unsupervised_lr_probability)
y_pred_unsupervised = np.argmax(
probability(theta, pi_y, pi, l_unsup_batch, s_unsup_batch, k, n_classes, continuous_mask).detach().numpy(),
1)
loss_3 = supervised_criterion(lr_model(X_unsup_batch), torch.tensor(y_pred_unsupervised))
loss_5 = log_likelihood_loss(theta, pi_y, pi, l_unsup_batch, s_unsup_batch, k, n_classes, continuous_mask)
prec_loss = precision_loss(theta, k, n_classes, a)
probs_graphical = probability(theta, pi_y, pi, l_batch, s_batch, k, n_classes, continuous_mask)
# print('probs_graphical',probs_graphical.shape)
probs_graphical = (probs_graphical.t() / probs_graphical.sum(1)).t()
probs_lr = torch.nn.Softmax()(lr_model(X_batch))
loss_6 = kl_divergence(probs_graphical, probs_lr)
if X_sup_batch.shape[0] != 0:
loss = loss_3 + loss_6 + loss_1 + loss_4 + loss_2 + loss_5
else:
loss = loss_3 + loss_6 + loss_2 + loss_5 # + loss_3 + loss_4
# loss = loss + prec_loss # loss_4#+loss_2 #+ loss_3# + prec_loss
# print(loss.item())
loss.backward()
optimizer.step()
y_pred = np.argmax(probability(theta, pi_y, pi, l_test, s_test, k, n_classes, continuous_mask).detach().numpy(), 1)
cur_gm_acc = accuracy_score(y_true_test, y_pred)
cur_gm_f1 = f1_score(y_true_test, y_pred, average="binary")
if (gm_acc < cur_gm_acc or gm_mac < cur_gm_f1):
gm_acc = cur_gm_acc
gm_mac = cur_gm_f1
highep = epoch
print("GM", lr_mac, lr_acc, epoch)
probs = torch.nn.Softmax()(lr_model(x_test)) # [:, 0]
# probs = torch.stack([probs, 1 - probs]).T
y_pred = np.argmax(probs.detach().numpy(), 1)
cur_lr_acc = accuracy_score(y_true_test, y_pred)
cur_lr_mac = f1_score(y_true_test, y_pred, average="binary")
if (lr_acc < cur_lr_acc or lr_mac < cur_lr_mac):
lr_acc = cur_lr_acc
lr_mac = cur_lr_mac
highep = epoch
print("LR", lr_mac, lr_acc, epoch)
print("Run: {} LR Epoch: {}\tMacro f1_score: {}".format(run, highep, lr_mac))
print("Run: {} LR Epoch: {}\t Accuracy: {}".format(run, highep, lr_acc))
print("Run: {} GM Epoch: {}\tMacro f1_score: {}".format(run, highep, gm_mac))
print("Run: {} GM Epoch: {}\t Accuracy: {}".format(run, highep, gm_acc))
lr_acc_list.append(lr_acc)
lr_mac_list.append(lr_mac)
gm_acc_list.append(gm_acc)
gm_mac_list.append(gm_mac)
print("Mean LR Macro f1_score: {}".format(np.array(lr_mac_list).mean()))
print("Mean LR Accuracy: {}".format(np.array(lr_acc_list).mean()))
print("Mean GM Macro f1_score: {}".format(np.array(gm_mac_list).mean()))
print("Mean Accuracy: {}".format(np.array(gm_acc_list).mean()))