-
Notifications
You must be signed in to change notification settings - Fork 0
/
dinclass_auto.py
325 lines (269 loc) · 12.5 KB
/
dinclass_auto.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
# Automated incremental learning models on all steps with collecting AUC-ROC and CE values per step
import os
import numpy as np
import pandas as pd
from sklearn.metrics import confusion_matrix
from sklearn import metrics
from keras import backend as K
from keras.models import Sequential, Model
from keras.layers import Bidirectional
from keras.layers import Input, Dense, LSTM
import tensorflow
from tensorflow.keras.optimizers import Adam
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID" # see issue #152
os.environ["CUDA_VISIBLE_DEVICES"] = ''
def train_test_split_week(week_test, updating = False):
df_indxs = ttab
ind_test = df_indxs[(df_indxs['last_week'] >= week_test) & (df_indxs['last_week'] < week_train)]
ind_test_2 = df_indxs[(df_indxs['last_week'] >= week_train) & (df_indxs['last_week'] < week_test_2)]
if updating:
ind_train = df_indxs[(df_indxs['last_week'] >= week_upd) & (df_indxs['last_week'] < week_test)]
else:
ind_train = df_indxs[(df_indxs['last_week'] >= week_test-L_win) & (df_indxs['last_week'] < week_test)] #!!!
return ind_test, ind_train, ind_test_2
class DataGenerator(tensorflow.keras.utils.Sequence):
def __init__(self, df, indexes, L_win, NCATS, batch_size):
self.data = df
self.batch_size = batch_size
self.ind = indexes
self.L_win = L_win
self.NCATS = NCATS
def __len__(self):
return int(np.floor(len(self.ind) / self.batch_size))
def __getitem__(self, idx):
batch_ind = self.ind[idx * self.batch_size:(idx + 1) * self.batch_size]
Ck = batch_ind[:, 0]
month = batch_ind[:, 1] - 1
ind_x = batch_ind[:, -(self.L_win + 1) : -1]
ind_y = batch_ind[:, -1]
X = self.data[ind_x, : ]
Y = self.data[ind_y, :]
Y = np.where(self.data[ind_y,:], 1, 0)
X = X.reshape(self.batch_size, self.L_win, self.NCATS)
Y = Y.reshape(self.batch_size, self.NCATS)
return [X, Ck, month], Y
def create_model():
inp = Input(shape=(L_win, NCATS))
inp_ck = Input(shape = (1, ))
inp_m = Input(shape = (1, ))
lay = LSTM(NFILTERS, return_sequences = True)(inp)
lay2 = LSTM(NFILTERS)(lay)
trg_clf = Dense(NCATS, activation = 'sigmoid')(lay2)
model_clf = Model(inputs = [inp, inp_ck, inp_m], outputs = trg_clf)
model_clf.compile(loss = 'binary_crossentropy', optimizer = OPTIM, metrics = ['accuracy'])
return model_clf
def answers_for_classes(classA, classB):
ind_test_A = ind_test[ind_test['id'].isin(classA)]
ind_test_B = ind_test[ind_test['id'].isin(classB)]
y_pred_A, y_true_A = get_answers_for_classes(ind_test_A)
y_pred_B, y_true_B = get_answers_for_classes(ind_test_B)
ind_cat = table.columns.get_loc(pred_cat) - 2
return y_pred_A[:, ind_cat], y_true_A[:, ind_cat], y_pred_B[:, ind_cat], y_true_B[:, ind_cat]
def get_answers_for_classes(ind_test):
model_RNN = create_model()
model_RNN.load_weights("LSTM.h5")
g_test = DataGenerator(table.values[:,2:], ind_test.values, L_win, NCATS, BATCH_SIZE)
y_pred = model_RNN.predict(g_test)
y_true = np.vstack([g_test[i][1] for i in range(len(g_test))])
del model_RNN
return y_pred, y_true
def train_and_predict(updating = False):
g_train = DataGenerator(table.values[:, 2:], ind_train.values, L_win, NCATS, BATCH_SIZE)
g_test = DataGenerator(table.values[:, 2:], ind_test.values, L_win, NCATS, BATCH_SIZE)
g_test2 = DataGenerator(table.values[:, 2:], ind_test_2.values, L_win, NCATS, BATCH_SIZE)
model_RNN = create_model()
if updating:
model_RNN.load_weights("LSTM.h5")
model_RNN.fit(g_train, validation_data = g_test, epochs = NB_EPOCH, verbose = 0)
model_RNN.save_weights('LSTM.h5')
y_pred = model_RNN.predict(g_test)
y_pred2 = model_RNN.predict(g_test2)
y_true = np.vstack([g_test[i][1] for i in range(len(g_test))])
ind_cat = table.columns.get_loc(pred_cat) - 2
del model_RNN
return y_pred[:, ind_cat], y_true[:, ind_cat], y_pred2[:, ind_cat]
def count_coefficient(y_pred, y_true, y_pred2):
ind_test_loc = ind_test.copy()
ind_test_loc2 = ind_test_2.copy()
#the first interval
ind_test_loc['target'] = table.values[ind_test_loc.values[:, -1], table.columns.get_loc(pred_cat)]
ind_test_loc['target'] = np.where(ind_test_loc['target'], 1, 0)
ind_test_loc['predicted_prob'] = np.zeros(len(ind_test_loc))
ind_test_loc.iloc[:len(y_pred), ind_test_loc.columns.get_loc('predicted_prob')] = y_pred
ind_test_loc = ind_test_loc[:len(y_pred)]
ind_test_loc['num'] = abs(ind_test_loc['target'] - ind_test_loc['predicted_prob'])
num = ind_test_loc.groupby('id')['num'].sum()
den_n = ind_test_loc.groupby('id')['target'].count()
CE = num / den_n
CE = 1 - CE
#the second interval
ind_test_loc2['target'] = table.values[ind_test_loc2.values[:, -1], table.columns.get_loc(pred_cat)]
ind_test_loc2['target'] = np.where(ind_test_loc2['target'], 1, 0)
ind_test_loc2['predicted_prob'] = np.zeros(len(ind_test_loc2))
ind_test_loc2.iloc[:len(y_pred2), ind_test_loc2.columns.get_loc('predicted_prob')] = y_pred2
ind_test_loc2 = ind_test_loc2[:len(y_pred2)]
ind_test_loc2['num'] = abs(ind_test_loc2['target'] - ind_test_loc2['predicted_prob'])
num2 = ind_test_loc2.groupby('id')['num'].sum()
den_n2 = ind_test_loc2.groupby('id')['target'].count()
CE2 = num2 / den_n2
CE2 = 1 - CE2
return CE, CE2
def predictability_classes(CE):
A=np.where(CE>np.median(CE))[0]
B=np.where(CE<np.median(CE))[0]
return A, B
def train_test_split_for_class_identification():
ind = np.arange(week_test, week_train)
ind_test = np.arange(week_train, week_test_2)
X_train = table[table['WEEK'].isin(ind)]
X_train = X_train[X_train['id'].isin(CE.keys())]
X_test = table[table['WEEK'].isin(ind_test)]
X_test = X_test[X_test['id'].isin(CE2.keys())]
test_ids = np.unique(X_test['id'])
return X_train, X_test, test_ids
def reshape_train_and_test(updating = False):
list_X, list_X_test = [], []
X_train, X_test, test_ids = train_test_split_for_class_identification()
un_ids = np.unique(X_train['id'])
Y = np.zeros(len(un_ids))
un_ids_test = np.unique(X_test['id'])
for c_id in un_ids:
cur = X_train[X_train['id'] == c_id][pred_cat].values
list_X.append(cur)
for i in range(len(un_ids)):
if un_ids[i] in A:
Y[i] = 0
elif un_ids[i] in B:
Y[i] = 1
for i in range(len(list_X)):
list_X[i] = list_X[i][-parameter:]
X = np.vstack(list_X)
X = X.reshape(len(list_X), n_timesteps, 1)
Y = pd.get_dummies(Y).values
Y = Y.reshape(len(list_X), 2)
for c_id in un_ids_test:
cur = X_test[X_test['id'] == c_id][pred_cat].values
list_X_test.append(cur)
for i in range(len(list_X_test)):
list_X_test[i] = list_X_test[i][-parameter:]
X_test_n = np.vstack(list_X_test)
X_test_n = X_test_n.reshape(len(list_X_test), n_timesteps, 1)
return X, Y, X_test_n, test_ids
def BiLSTM_model(updating = False):
X, Y, X_test_n, test_ids = reshape_train_and_test(updating = updating)
model = Sequential()
model.add(Bidirectional(LSTM(BiNFILTERS, input_shape=(n_timesteps, 1),
return_sequences = False)))
model.add(Dense(2, activation = 'softmax'))
model.compile(loss = 'categorical_crossentropy', optimizer = 'adam', metrics = ['acc'])
X = X.astype('float32')
Y = Y.astype('float32')
X_test_n = X_test_n.astype('float32')
if updating:
model(np.zeros((1,*X.shape[1:])))
model.load_weights('BiLSTM.h5')
model.fit(X, Y, epochs = BiNB_EPOCHS, batch_size = BiBATCH_SIZE, verbose=0)
model.save_weights('BiLSTM.h5')
yhat = model.predict(X_test_n).argmax(1)
pred_A, pred_B = [], []
for i in range(len(yhat)):
if (yhat[i] == 0):
pred_A.append(test_ids[i])
else:
pred_B.append(test_ids[i])
del model
return pred_A, pred_B
def data_for_matrix(y_pred,threshold):
y_pred_matrix=np.zeros(shape=y_pred.shape)
for i in range(0,len(y_pred)):
if y_pred[i]>threshold:
y_pred_matrix[i]=1
return y_pred_matrix
def infer(updating = False):
global y_pred, y_true, y_pred2, CE, CE2, A, B
y_pred, y_true, y_pred2 = train_and_predict(updating = updating)
CE, CE2 = count_coefficient(y_pred, y_true, y_pred2)
A, B = predictability_classes(CE)
y_pred_A, y_true_A, y_pred_B, y_true_B = answers_for_classes(A, B)
pred_A, pred_B = BiLSTM_model(updating = updating)
y_pred_est_A, y_true_est_A, y_pred_est_B, y_true_est_B = answers_for_classes(pred_A, pred_B)
return y_pred, y_true, y_pred_A, y_true_A, y_pred_B, y_true_B, y_pred_est_A, y_true_est_A, y_pred_est_B, y_true_est_B
L_win = 4
delta = 2
pred_cat = '60'
n_timesteps = 4
parameter=4
num_month_client=3
NFILTERS = 64
lr = 0.001
BATCH_SIZE = 64
NB_EPOCH = 10
BiNFILTERS = 20
BiNB_EPOCHS = 30
BiBATCH_SIZE = 32
OPTIM = Adam(learning_rate=lr)
datadir='data/'
setname='D1'
table=pd.read_csv(datadir+setname+'table.csv')
ttab=pd.read_csv(datadir+setname+'indtab.csv').astype(int)
week_test = 6 - delta
week_train = 10 - delta
week_test_2 = 14 - delta
NCATS = table.shape[1] - 2
colors = ['k','g','r','g', 'r']
linestyles = ['-', '-', '-', '-', '-']
accA, accB, accTot=[],[],[]
aucA, aucB, aucTot=[],[],[]
ce=[]
for step in range(20):
K.clear_session()
week_test += delta
week_train += delta
week_test_2 += delta
week_upd = week_test - delta
if week_test_2 > ttab.last_week.max():
print('\tNo more data. Game over!')
break
print('step # %i ===week_test:%i week_train:%i week_test_2:%i week_upd:%i ========='%(step,
week_test, week_train, week_test_2, week_upd))
if step==0:
upd=False
else:
upd=True
ind_test, ind_train, ind_test_2 = train_test_split_week(week_test, updating = upd)
y_pred_LSTM, y_true_LSTM, y_pred_A, y_true_A, y_pred_B, y_true_B, y_pred_est_A, y_true_est_A, y_pred_est_B, y_true_est_B = infer(updating = upd)
y_pred_est_A1=data_for_matrix(y_pred_est_A,np.median(CE))
confusion_matrix_est_A=confusion_matrix(y_true_est_A,y_pred_est_A1)
accuracy1 = (confusion_matrix_est_A[0][0]+confusion_matrix_est_A[1][1])/(confusion_matrix_est_A[0][0]+confusion_matrix_est_A[1][0]+confusion_matrix_est_A[0][1]+confusion_matrix_est_A[1][1])
print('Accuracy for class A after %i learning is : %.3f'%(step+1, accuracy1))
y_pred_est_B1=data_for_matrix(y_pred_est_B,np.median(CE))
confusion_matrix_est_B=confusion_matrix(y_true_est_B,y_pred_est_B1)
accuracy2 = (confusion_matrix_est_B[0][0]+confusion_matrix_est_B[1][1])/(confusion_matrix_est_B[0][0]+confusion_matrix_est_B[1][0]+confusion_matrix_est_B[0][1]+confusion_matrix_est_B[1][1])
print('Accuracy for class B after %i learning is : %.3f'%(step+1, accuracy2))
accuracy_classes_numenator = (confusion_matrix_est_B[0][0]+confusion_matrix_est_B[1][1]+confusion_matrix_est_A[0][0]+confusion_matrix_est_A[1][1])
accuracy_classes_denumenator = (confusion_matrix_est_B[0][0]+confusion_matrix_est_B[1][0]+confusion_matrix_est_B[0][1]+confusion_matrix_est_B[1][1] + confusion_matrix_est_A[0][0]+confusion_matrix_est_A[1][0]+confusion_matrix_est_A[0][1]+confusion_matrix_est_A[1][1])
accuracy_classes= accuracy_classes_numenator/accuracy_classes_denumenator
print('Accuracy after %i learning is : %.3f'%(step+1, accuracy_classes))
total_true_est = np.concatenate((y_true_est_A,y_true_est_B), axis=0)
total_pred_est = np.concatenate((y_pred_est_A,y_pred_est_B), axis=0)
fpr_A, tpr_A, thresholds_A = metrics.roc_curve(y_true_est_A, y_pred_est_A, pos_label=1)
roc_auc_A = metrics.auc(fpr_A, tpr_A)
fpr_B, tpr_B, thresholds_B = metrics.roc_curve(y_true_est_B, y_pred_est_B, pos_label=1)
roc_auc_B = metrics.auc(fpr_B, tpr_B)
fpr, tpr, thresholds = metrics.roc_curve(total_true_est, total_pred_est, pos_label=1)
roc_auc = metrics.auc(fpr, tpr)
print('AUC unpredictable class',round(roc_auc_B,3))
print('AUC predictable class',round(roc_auc_A,3))
print('AUC both classes',round(roc_auc,3))
accA.append(accuracy1)
accB.append(accuracy2)
accTot.append(accuracy_classes)
aucA.append(roc_auc_A)
aucB.append(roc_auc_B)
aucTot.append(roc_auc)
ce.append(np.median(CE))
'''save collected values to DataFrame'''
acc_df=pd.DataFrame({'step':np.arange(len(accA))+1,
'accA':accA, 'accB':accB, 'accTot':accTot,
'aucA':aucA, 'aucB':aucB, 'aucTot':aucTot,'CE':ce})
acc_df.to_csv(datadir+setname+'increm.csv', index=False)