-
Notifications
You must be signed in to change notification settings - Fork 2
/
06_predict_25.py
407 lines (351 loc) · 16 KB
/
06_predict_25.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
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
#=======================================================================================================================
# This file generates firsr level predictions using trained models saved in the folder weights/
# predictions will be saved in wdir/submits folder
# Predict on window size (2.5*segment length), where segment length = total counts in test file / 30,
# scan for 200 windows
# @ Shaikat Galib / [email protected] / 25/04/2019
#=======================================================================================================================
import os
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
from collections import Counter
import sys
import pandas as pd
import numpy as np
np.random.seed(203)
from tqdm import tqdm
from sklearn.preprocessing import RobustScaler
from sklearn.externals import joblib
import warnings
warnings.filterwarnings("ignore")
from keras.layers import LSTM
from keras.models import Model ,load_model
try:
import cPickle as pickle
except BaseException:
import pickle
#=======================================================================================================================
# bash test.sh /data/testing/ solution.csv
########################################################################################################################
test_folder = '/data_small/testing/'
solution_fn = 'solution.csv'
wdata_dir = '/wdata/'
if len(sys.argv) > 1:
test_folder = sys.argv[1]
solution_fn = sys.argv[2]
files = sorted(os.listdir(test_folder))
#=======================================================================================================================
# folder name to save submits
expt_name = 'ann'
# weight directory
expt_name1 = 'ANN_CNN'
# expt_name2 = 'lstm'
# expt_name3 = 'lgb'
# #=======================================================================================================================
# read data
#=======================================================================================================================
#df_test = pd.read_csv(data_dir + 'submittedAnswers.csv')
df_feat = pd.read_csv(wdata_dir + 'train_feature_bin_30_slice.csv')
# =======================================================================================================================
# segment multiplication factor
seg_mul = 1.25
#======================================================================================================================
# make bins like train
energy_bin_size = 30
energy_bins = np.arange(0, 3000, energy_bin_size)
num_windows = 200
# Make submission directories
sub_dir = wdata_dir + 'submits/' + expt_name + '/'
if not os.path.exists(sub_dir):
os.makedirs(sub_dir)
# =======================================================================================================================
# function for finding source time by calculating mid segment of deected segments
def smooth(y, box_pts):
box = np.ones(box_pts) / box_pts
y_smooth = np.convolve(y, box, mode='same')
return y_smooth
# # lstm model that runs on cpu
# def lstm_model():
# inp = Input(shape=(99, 1))
# a = LSTM(100, return_sequences=True, recurrent_activation='sigmoid')(inp)
# a = LSTM(20, recurrent_activation='sigmoid')(a)
# a = Dense(128, activation="relu", kernel_initializer="uniform")(a)
# output = Dense(7, activation="softmax", kernel_initializer="uniform")(a)
# model = Model(inp, output)
# return model
#
# def ann_model():
# inp = Input(shape=(len(X[0]), 1))
#
# a = Conv1D(64, 5, activation="relu", kernel_initializer="uniform", )(inp)
# a = BatchNormalization()(a)
# a = Conv1D(64, 5, activation="relu", kernel_initializer="uniform", )(a)
# a = BatchNormalization()(a)
# max_pool = GlobalMaxPool1D()(a)
#
# b = Flatten()(inp)
# ab = concatenate([ max_pool, b])
#
# a = Dense(128, activation="relu", kernel_initializer="uniform")(ab)
# a = Dropout(0.5)(a)
# a = Dense(128, activation="relu", kernel_initializer="uniform")(a)
# output = Dense(7, activation="softmax", kernel_initializer="uniform")(a)
# model = Model(inp, output)
#
# return model
#=======================================================================================================================
def make_features(energy):
out = pd.cut(energy, bins=energy_bins, include_lowest=True)
counts = out.value_counts(sort=False)
np_counts = np.array(counts.values, dtype=np.float32)
np_counts = np_counts / np.sum(np_counts)
# ===================================================================================================================
# peak to compton ratio feature
# ===================================================================================================================
count1 = np_counts[1] / np.sum(np_counts[0:1])
count2 = np_counts[2] / np.sum(np_counts[0:2])
count3 = np_counts[3] / np.sum(np_counts[0:2])
count4 = np_counts[4] / np.sum(np_counts[0:3])
count5 = np_counts[5] / np.sum(np_counts[0:4])
count6 = np_counts[6] / np.sum(np_counts[0:5])
count7 = np_counts[7] / np.sum(np_counts[0:6])
count8 = np_counts[8] / np.sum(np_counts[0:7])
count9 = np_counts[9] / np.sum(np_counts[0:8])
count10 = np_counts[10] / np.sum(np_counts[0:9])
count11 = np_counts[11] / np.sum(np_counts[0:10])
count12 = np_counts[12] / np.sum(np_counts[0:11])
count13 = np_counts[13] / np.sum(np_counts[0:12])
count14 = np_counts[14] / np.sum(np_counts[0:13])
count15 = np_counts[15] / np.sum(np_counts[0:14])
count16 = np_counts[16] / np.sum(np_counts[0:15])
count17 = np_counts[17] / np.sum(np_counts[0:16])
count18 = np_counts[18] / np.sum(np_counts[0:17])
count19 = np_counts[19] / np.sum(np_counts[0:18])
count20 = np_counts[20] / np.sum(np_counts[0:19])
count21 = np_counts[21] / np.sum(np_counts[0:20])
count22 = np_counts[87] / np.sum(np_counts[0:86])
np_counts_peaks = np.array(
[count1, count2, count3, count4, count5, count6, count7, count8, count9, count10,
count11, count12, count13, count14, count15, count16, count17, count18, count19, count20,
count21, count22]).T
# ===================================================================================================================
# peak to peak ratio feature
# ===================================================================================================================
# HEU
ratio1 = np_counts[0] / (np_counts[17] + np_counts[19])
ratio2 = np_counts[3] / (np_counts[17] + np_counts[19])
ratio3 = np_counts[6] / (np_counts[17] + np_counts[19])
ratio4 = np_counts[87] / (np_counts[17] + np_counts[19])
# WGPu
ratio5 = np_counts[1] / (np_counts[12] + np_counts[13])
ratio6 = np_counts[2] / (np_counts[12] + np_counts[13])
ratio7 = np_counts[3] / (np_counts[12] + np_counts[13])
ratio8 = np_counts[6] / (np_counts[12] + np_counts[13])
ratio9 = np_counts[21] / (np_counts[12] + np_counts[13])
# I-131
ratio10 = np_counts[1] / np_counts[12]
ratio11 = np_counts[2] / np_counts[12]
ratio12 = np_counts[6] / np_counts[12]
ratio13 = np_counts[9] / np_counts[12]
ratio14 = np_counts[21] / np_counts[12]
# Tc-99m
ratio15 = np_counts[0] / np_counts[4]
ratio16 = np_counts[1] / np_counts[4]
ratio17 = np_counts[10] / np_counts[4]
# HEU + Tc-99m
ratio18 = np.sum(np_counts[0:7]) / np.sum(np_counts[0:21])
ratio19 = ratio1 / ratio15
ratio20 = ratio2 / ratio15
ratio21 = ratio3 / ratio15
ratio22 = ratio4 / ratio15
ratio23 = ratio1 / ratio16
ratio24 = ratio2 / ratio16
ratio25 = ratio3 / ratio16
ratio26 = ratio4 / ratio16
ratio27 = ratio1 / ratio17
ratio28 = ratio2 / ratio17
ratio29 = ratio3 / ratio17
ratio30 = ratio4 / ratio17
np_ratio_peaks = np.array(
[ratio1, ratio2, ratio3, ratio4, ratio5, ratio6, ratio7, ratio8, ratio9, ratio10,
ratio11, ratio12, ratio13, ratio14, ratio15, ratio16, ratio17, ratio18, ratio19, ratio20,
ratio21, ratio22, ratio23, ratio24, ratio25, ratio26, ratio27, ratio28, ratio29, ratio30]).T
feats = np.concatenate([np_counts, np_counts_peaks, np_ratio_peaks], axis=0)
return feats
# =======================================================================================================================
# Empty list list tostore results
answer1_id = []
answer1_fn = []
answer1_tm = []
answer2_id = []
answer2_fn = []
answer2_tm = []
answer3_id = []
answer3_fn = []
answer3_tm = []
# =======================================================================================================================
# stats from train set
# =======================================================================================================================
#target = df_train.iloc[:, -1]
x_trn = df_feat.iloc[:,1:-1]
# scale parameters from train
X = x_trn.values
where_are_NaNs = np.isnan(X)
where_are_infs = np.isinf(X)
X[where_are_NaNs] = 0
X[where_are_infs] = 0
scaler = RobustScaler()
scaler.fit(X)
# =======================================================================================================================
# iterate over each file from test set
# =======================================================================================================================
model1 = load_model('weights/{}/model_{}.hdf5'.format(expt_name1, 0))
model2 = load_model('weights/{}/model_{}.hdf5'.format(expt_name1, 1))
model3 = load_model('weights/{}/model_{}.hdf5'.format(expt_name1, 2))
model4 = load_model('weights/{}/model_{}.hdf5'.format(expt_name1, 3))
model5 = load_model('weights/{}/model_{}.hdf5'.format(expt_name1, 4))
print('model loaded')
# model2 = lstm_model()
# model2.load_weights('weights/{}/model_{}.hdf5'.format(expt_name2, fold_))
# with open('weights/{}/model_{}.pkl'.format(expt_name3, fold_), 'rb') as fin:
# model3 = pickle.load(fin)
for i, id in enumerate(tqdm(files)):
id = os.path.splitext(id)[0]
df = pd.read_csv(test_folder + '{}.csv'.format(id))
time = df[df.columns[0]]
energy = df[df.columns[1]]
length = len(time)
df['time_cumsum'] = np.array(time.cumsum(), dtype=np.float32)
# divide test file into 30 segments
seg_width = int(length / 30)
if seg_width > 10000: seg_width = 10000
if seg_width < 2000: seg_width = 2000
# Create equally spaced slices
f = lambda m, n: [k * n // m + n // (2 * m) for k in range(m)]
_idxs = f(num_windows, length) # 200 slices
# find 30 sec index
df_sort = df.ix[(df['time_cumsum'] - 30000000).abs().argsort()[:2]]
idx30 = df_sort.index.tolist()[0]
seg_idxs = [k for k in _idxs if k >= idx30]
batch_inp_counts = np.zeros((len(seg_idxs), len(X[0]), 1))
for s, j in enumerate(seg_idxs):
###################################################
start = int(j - seg_width * seg_mul)
end = int(j + seg_width * seg_mul)
if j - seg_width * seg_mul < 0: start = 0
if j + seg_width * seg_mul > length: end = length
###################################################
seg = df[start:end]
energy = seg[seg.columns[1]]
features = make_features(energy)
where_are_NaNs = np.isnan(features)
where_are_infs = np.isinf(features)
features[where_are_NaNs] = 0
features[where_are_infs] = 0
##################################################
# df_counts = np_counts.reshape(1,99,1)
inp_counts = features
inp_counts = inp_counts.reshape(len(features), 1)
batch_inp_counts[s, :, ] = inp_counts
##################################################
batch_inp_counts = np.squeeze(batch_inp_counts)
# ==========================================================
# scale input
#lgb
scaled_train_X = scaler.transform(batch_inp_counts)
# # lstm
# X_lstm = scaled_train_X[:, 0:99]
# X_lstm = X_lstm.reshape(len(seg_idxs), 99, 1)
# ANN
X = scaled_train_X.reshape(len(seg_idxs), len(X[0]), 1)
# ========================================================
# ANN
pred1 = model1.predict(X, batch_size=len(seg_idxs))
pred2 = model2.predict(X, batch_size=len(seg_idxs))
pred3 = model3.predict(X, batch_size=len(seg_idxs))
pred4 = model4.predict(X, batch_size=len(seg_idxs))
pred5 = model5.predict(X, batch_size=len(seg_idxs))
# # lstm
# pred2 = model2.predict(X_lstm, batch_size=len(seg_idxs))
# # lgb
# pred3 = model3.predict(scaled_train_X)
pred = pred1 + pred2 + pred3 + pred4 + pred5
p = pred.argmax(axis=1)
# update dataframe
for m, n in enumerate(seg_idxs):
df.loc[int(n), 'label'] = p[m]
df = df[np.isfinite(df['label'])]
# ===================================================================================================================
# Count the frequency of non-zero preds
try:
preds_nonzero = [x for x in p if x > 0]
most_common, freq_most_common = Counter(preds_nonzero).most_common(1)[0]
except:
freq_most_common = 0
# ===================================================================================================================
# If frequency of non-zero predictions are greater than a threshold, then it's positive
if freq_most_common > 7:
# option 1: replace all non zero with most frequent
# df['flag'] = np.where((df['label'] > 0), most_common, 0)
# option 2: everything zero except most frequent
df['flag'] = np.where((df['label'] != most_common), 0, most_common)
preds_ = df['flag'].tolist()
preds_nonzero = [x for x in preds_ if x > 0]
most_common_, freq_most_common_ = Counter(preds_nonzero).most_common(1)[0]
df['grad'] = smooth(df['flag'], freq_most_common_)
nearest_time = df.loc[(df.grad.idxmax(), 'time_cumsum')]
answer1_fn.append(id)
answer1_id.append(most_common)
answer1_tm.append(nearest_time / 1000000)
else:
answer1_fn.append(id)
answer1_id.append(0)
answer1_tm.append(0)
if freq_most_common > 5:
df['flag'] = np.where((df['label'] != most_common), 0, most_common)
preds_ = df['flag'].tolist()
preds_nonzero = [x for x in preds_ if x > 0]
most_common_, freq_most_common_ = Counter(preds_nonzero).most_common(1)[0]
df['grad'] = smooth(df['flag'], freq_most_common_)
nearest_time = df.loc[(df.grad.idxmax(), 'time_cumsum')]
answer2_fn.append(id)
answer2_id.append(most_common)
answer2_tm.append(nearest_time / 1000000)
else:
answer2_fn.append(id)
answer2_id.append(0)
answer2_tm.append(0)
if freq_most_common > 3:
df['flag'] = np.where((df['label'] != most_common), 0, most_common)
preds_ = df['flag'].tolist()
preds_nonzero = [x for x in preds_ if x > 0]
most_common_, freq_most_common_ = Counter(preds_nonzero).most_common(1)[0]
df['grad'] = smooth(df['flag'], freq_most_common_)
nearest_time = df.loc[(df.grad.idxmax(), 'time_cumsum')]
answer3_fn.append(id)
answer3_id.append(most_common)
answer3_tm.append(nearest_time / 1000000)
else:
answer3_fn.append(id)
answer3_id.append(0)
answer3_tm.append(0)
# =======================================================================================================================
sub1 = pd.DataFrame()
sub1["RunID"] = answer1_fn
sub1["SourceID"] = answer1_id
sub1["SourceTime"] = answer1_tm
sub1 = sub1.sort_values('RunID')
sub1.to_csv(sub_dir + "solution_{}_th7_seg{}_test.csv".format(expt_name, seg_mul), index=False)
sub2 = pd.DataFrame()
sub2["RunID"] = answer2_fn
sub2["SourceID"] = answer2_id
sub2["SourceTime"] = answer2_tm
sub2 = sub2.sort_values('RunID')
sub2.to_csv(sub_dir + "solution_{}_th5_seg{}_test.csv".format(expt_name, seg_mul), index=False)
sub3 = pd.DataFrame()
sub3["RunID"] = answer3_fn
sub3["SourceID"] = answer3_id
sub3["SourceTime"] = answer3_tm
sub3 = sub3.sort_values('RunID')
sub3.to_csv(sub_dir + "solution_{}_th3_seg{}_test.csv".format(expt_name, seg_mul), index=False)
print('done')