-
Notifications
You must be signed in to change notification settings - Fork 1
/
fast_solutionMod#14.py
367 lines (324 loc) · 17 KB
/
fast_solutionMod#14.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
from datetime import datetime
from math import log, exp, sqrt
# parameters #################################################################
train = 'train.csv' # path to training file
label = 'trainLabels.csv' # path to label file of training data
test = 'test.csv' # path to testing file
solution = 14
monitor = open('diag'+str(solution)+'.out','w')
D = 2 ** 24 # number of weights use for each model, we have 32 of them
alpha = .1 # learning rate for sgd optimization
def is_number(s):
try:
float(s)
return True
except ValueError:
return False
# function, generator definitions ############################################
# A. x, y generator
# INPUT:
# path: path to train.csv or test.csv
# label_path: (optional) path to trainLabels.csv
# YIELDS:
# ID: id of the instance (can also acts as instance count)
# x: a list of indices that its value is 1
# y: (if label_path is present) label value of y1 to y33
def data(path, label_path=None):
for t, line in enumerate(open(path)):
# initialize our generator
if t == 0:
# create a static x,
# so we don't have to construct a new x for every instance
x = [0] * (146 + 46 + 1 + 7 + 2 + 7 + 10 + 7 + 7 + 7 + 4 + 5 + 14 + 60 + 20 + 5 + 60 + 20 + 5)
if label_path:
label = open(label_path)
label.readline() # we don't need the headers
continue
# parse x
row = line.rstrip().split(',')
for m, feat in enumerate(line.rstrip().split(',')):
if m == 0:
ID = int(feat)
else:
# one-hot encode everything with hash trick
# categorical: one-hotted
# boolean: ONE-HOTTED
# numerical: ONE-HOTTED!
# note, the build in hash(), although fast is not stable,
# i.e., same value won't always have the same hash
# on different machines
if is_number(feat):
feat=str(round(float(feat),1))
x[m] = abs(hash(str(m) + '_' + feat)) % D
hash_cols = [3,4,34,35,61,64,65,91,94,95]
t = 145
for i in xrange(10):
for j in xrange(i+1,10):
t += 1
x[t] = abs(hash(row[hash_cols[i]]+"_x_"+row[hash_cols[j]])) % D
#+1
nul = 0 if row[1] == "" else 1
x[192] = abs(hash(str(nul) + '_nul')) % D
#+7
hesh = 'a' if float(row[5]) == 0 else ('b' if float(row[5]) >= 1 else 'c')
x[193]= abs(hash('i5_' + hesh)) % D
hesh = 'a' if float(row[6]) == 0 else ('b' if float(row[6]) >= 1 else 'c')
x[194]= abs(hash('i6_' + hesh)) % D
hesh = 'a' if float(row[7]) == 0 else ('b' if float(row[7]) == 1 else 'c')
x[195]= abs(hash('i7_' + hesh)) % D
hesh = 'a' if float(row[8]) == 0 else ('b' if float(row[8]) >= 1 else 'c')
x[196]= abs(hash('i8_' + hesh)) % D
hesh = 'a' if float(row[9]) == 0 else ('b' if float(row[9]) >= 1 else 'c')
x[197]= abs(hash('i9_' + hesh)) % D
hesh = 0 if float(row[5]) == 0 and float(row[6]) == 0 and float(row[7]) == 0 and float(row[8]) == 0 and float(row[9]) else 1
x[198]= abs(hash('i56789_' + str(hesh))) % D
hesh = round(float(row[5]) + float(row[6]) + float(row[7]) + float(row[8]) + float(row[9]),2)
x[199]= abs(hash('i56789_sum' + str(round(float(hesh),2)))) % D
#+2
yesno_cols = [1,2,10,11,12,13,14,24,25,26,30,31,32,33,41,42,43,44,45,55,56,57,62,63,71,72,73,74,75,85,86,87,92,93,101,102,103,104,105,115,116,117,126,127,128,129,130,140,142]
yesno_hash = ""
for i in xrange(len(yesno_cols)):
yesno_hash += row[yesno_cols[i]] + "_x_"
x[200] = abs(hash(yesno_hash)) % D
nul = 0 if row[5] == 0 else 1
x[201] = abs(hash(str(nul) + '5_nul')) % D
#+7
hesh = 'a' if float(row[36]) == 0 else ('b' if float(row[36]) >= 1 else 'c')
x[202]= abs(hash('i36_' + hesh)) % D
hesh = 'a' if float(row[37]) == 0 else ('b' if float(row[37]) >= 1 else 'c')
x[203]= abs(hash('i37_' + hesh)) % D
hesh = 'a' if float(row[38]) == 0 else ('b' if float(row[38]) == 1 else 'c')
x[204]= abs(hash('i38_' + hesh)) % D
hesh = 'a' if float(row[39]) == 0 else ('b' if float(row[39]) >= 1 else 'c')
x[205]= abs(hash('i39_' + hesh)) % D
hesh = 'a' if float(row[40]) == 0 else ('b' if float(row[40]) >= 1 else 'c')
x[206]= abs(hash('i40_' + hesh)) % D
hesh = 0 if float(row[36]) == 0 and float(row[37]) == 0 and float(row[38]) == 0 and float(row[39]) == 0 and float(row[40]) else 1
x[207]= abs(hash('i3637383940_' + str(hesh))) % D
hesh = round(float(row[36]) + float(row[37]) + float(row[38]) + float(row[39]) + float(row[40]),2)
x[208]= abs(hash('i3637383940__sum' + str(round(float(hesh),2)))) % D
#+10
hesh = 'a' if float(row[20]) == 0 else ('b' if float(row[20]) > 0 else 'c')
x[209]= abs(hash('i20_' + hesh)) % D
nul = 1 if float(row[21]) == 1 else 0
x[210] = abs(hash(str(nul) + '21_nul')) % D
hesh = 'a' if float(row[51]) == 0 else ('b' if float(row[51]) > 0 else 'c')
x[211]= abs(hash('i51_' + hesh)) % D
nul = 1 if float(row[52]) == 1 else 0
x[212] = abs(hash(str(nul) + '52_nul')) % D
hesh = 'a' if float(row[81]) == 0 else ('b' if float(row[81]) > 0 else 'c')
x[213]= abs(hash('i81_' + hesh)) % D
nul = 1 if float(row[82]) == 1 else 0
x[214] = abs(hash(str(nul) + '82_nul')) % D
hesh = 'a' if float(row[111]) == 0 else ('b' if float(row[111]) > 0 else 'c')
x[215]= abs(hash('i111_' + hesh)) % D
nul = 1 if float(row[112]) == 1 else 0
x[216] = abs(hash(str(nul) + '112_nul')) % D
hesh = 'a' if float(row[136]) == 0 else ('b' if float(row[136]) > 0 else 'c')
x[217]= abs(hash('i136_' + hesh)) % D
nul = 1 if float(row[137]) == 1 else 0
x[218] = abs(hash(str(nul) + '137_nul')) % D
#+7
hesh = 'a' if float(row[66]) == 0 else ('b' if float(row[66]) >= 1 else 'c')
x[219]= abs(hash('i66_' + hesh)) % D
hesh = 'a' if float(row[67]) == 0 else ('b' if float(row[67]) >= 1 else 'c')
x[220]= abs(hash('i67_' + hesh)) % D
hesh = 'a' if float(row[68]) == 0 else ('b' if float(row[68]) == 1 else 'c')
x[221]= abs(hash('i68_' + hesh)) % D
hesh = 'a' if float(row[69]) == 0 else ('b' if float(row[69]) >= 1 else 'c')
x[222]= abs(hash('i69_' + hesh)) % D
hesh = 'a' if float(row[70]) == 0 else ('b' if float(row[70]) >= 1 else 'c')
x[223]= abs(hash('i70_' + hesh)) % D
hesh = 0 if float(row[66]) == 0 and float(row[67]) == 0 and float(row[68]) == 0 and float(row[69]) == 0 and float(row[70]) else 1
x[224]= abs(hash('i6667686970_' + str(hesh))) % D
hesh = round(float(row[66]) + float(row[67]) + float(row[68]) + float(row[69]) + float(row[70]),2)
x[225]= abs(hash('i6667686970__sum' + str(round(float(hesh),2)))) % D
#+7
hesh = 'a' if float(row[96]) == 0 else ('b' if float(row[96]) >= 1 else 'c')
x[226]= abs(hash('i96_' + hesh)) % D
hesh = 'a' if float(row[97]) == 0 else ('b' if float(row[97]) >= 1 else 'c')
x[227]= abs(hash('i97_' + hesh)) % D
hesh = 'a' if float(row[98]) == 0 else ('b' if float(row[98]) == 1 else 'c')
x[228]= abs(hash('i98_' + hesh)) % D
hesh = 'a' if float(row[99]) == 0 else ('b' if float(row[99]) >= 1 else 'c')
x[229]= abs(hash('i99_' + hesh)) % D
hesh = 'a' if float(row[100]) == 0 else ('b' if float(row[100]) >= 1 else 'c')
x[230]= abs(hash('i100_' + hesh)) % D
hesh = 0 if float(row[96]) == 0 and float(row[97]) == 0 and float(row[98]) == 0 and float(row[99]) == 0 and float(row[100]) else 1
x[231]= abs(hash('i96979899100_' + str(hesh))) % D
hesh = round(float(row[96]) + float(row[97]) + float(row[98]) + float(row[99]) + float(row[100]),2)
x[232]= abs(hash('i96979899100__sum' + str(round(float(hesh),2)))) % D
#+7
hesh = 'a' if float(row[121]) == 0 else ('b' if float(row[121]) >= 1 else 'c')
x[233]= abs(hash('i121_' + hesh)) % D
hesh = 'a' if float(row[122]) == 0 else ('b' if float(row[122]) >= 1 else 'c')
x[234]= abs(hash('i122_' + hesh)) % D
hesh = 'a' if float(row[123]) == 0 else ('b' if float(row[123]) == 1 else 'c')
x[235]= abs(hash('i123_' + hesh)) % D
hesh = 'a' if float(row[124]) == 0 else ('b' if float(row[124]) >= 1 else 'c')
x[236]= abs(hash('i124_' + hesh)) % D
hesh = 'a' if float(row[124]) == 0 else ('b' if float(row[124]) >= 1 else 'c')
x[237]= abs(hash('i124_' + hesh)) % D
hesh = 0 if float(row[121]) == 0 and float(row[122]) == 0 and float(row[123]) == 0 and float(row[124]) == 0 and float(row[125]) else 1
x[238]= abs(hash('i121122123124125_' + str(hesh))) % D
hesh = round(float(row[121]) + float(row[122]) + float(row[123]) + float(row[124]) + float(row[125]),2)
x[239]= abs(hash('121122123124125__sum' + str(round(float(hesh),2)))) % D
#+4
x[240] = abs(hash('i2223_' + str(round(float(row[22]),2)) + '_' + str(round(float(row[23]),2)))) % D
x[241] = abs(hash('i8384_' + str(round(float(row[83]),2)) + '_' + str(round(float(row[84]),2)))) % D
x[242] = abs(hash('i113114_' + str(round(float(row[113]),2)) + '_' + str(round(float(row[114]),2)))) % D
x[243] = abs(hash('i138139_' + str(round(float(row[138]),2)) + '_' + str(round(float(row[139]),2)))) % D
#+5
hesh = 1 if round(float(row[22]),2) > round(float(row[23]),2) else (-1 if round(float(row[23]),2) > round(float(row[22]),2) else 0)
x[244] = abs(hash('i2223_equal_' + str(hesh))) % D
hesh = 1 if round(float(row[53]),2) > round(float(row[54]),2) else (-1 if round(float(row[54]),2) > round(float(row[53]),2) else 0)
x[245] = abs(hash('i5354_equal_' + str(hesh))) % D
hesh = 1 if round(float(row[83]),2) > round(float(row[84]),2) else (-1 if round(float(row[84]),2) > round(float(row[83]),2) else 0)
x[246] = abs(hash('i8384_equal_' + str(hesh))) % D
hesh = 1 if round(float(row[113]),2) > round(float(row[114]),2) else (-1 if round(float(row[114]),2) > round(float(row[113]),2) else 0)
x[247] = abs(hash('i113114_equal_' + str(hesh))) % D
hesh = 1 if round(float(row[138]),2) > round(float(row[139]),2) else (-1 if round(float(row[139]),2) > round(float(row[138]),2) else 0)
x[248] = abs(hash('i138139_equal_' + str(hesh))) % D
#+14
x[249]= abs(hash('i1_i2_' + row[1] + row[2])) % D
x[250]= abs(hash('i10_i11_i12_i13_i14_' + row[10] + row[11] + row[12] + row[13] + row[14])) % D
x[251]= abs(hash('i24_i25_i26_' + row[24] + row[25] + row[26])) % D
x[252]= abs(hash('i30_i31_i31_i33_' + row[30] + row[31] + row[32] + row[33])) % D
x[253]= abs(hash('i41_i42_i43_i44_i45_' + row[41] + row[42] + row[43] + row[44] + row[45])) % D
x[254]= abs(hash('i55_i56_i57_' + row[55] + row[56] + row[57])) % D
x[255]= abs(hash('i62_i63_' + row[62] + row[63])) % D
x[256]= abs(hash('i71_i72_i73_i74_i75_' + row[71] + row[72] + row[73] + row[74] + row[75])) % D
x[257]= abs(hash('i85_i86_i87_' + row[85] + row[86] + row[87])) % D
x[258]= abs(hash('i92_i93_' + row[92] + row[93])) % D
x[259]= abs(hash('i101_i102_i103_i104_i105_' + row[101] + row[102] + row[103] + row[104] + row[105])) % D
x[260]= abs(hash('i115_i116_i117_' + row[115] + row[116] + row[117])) % D
x[261]= abs(hash('i126_i127_i128_i129_i130_' + row[126] + row[127] + row[128] + row[129] + row[130])) % D
x[262]= abs(hash('i140_i141_i142_' + row[140] + row[141] + row[142])) % D
t = 262
#+60+20+5
num_cols = [5,6,7,8,9,16,19,21,22,23,28,29,36,37,38,39,40,47,50,52,53,54,59,60,66,67,68,69,70,77,80,82,83,84,89,90,96,97,98,99,100,107,110,112,113,114,119,120,121,122,123,124,125,132,135,137,138,139,144,145]
int_cols = [15,17,18,46,48,49,53,58,76,78,79,88,106,108,109,118,131,133,134,143]
zero_cols = [20,51,81,111,136]
for i in xrange(len(num_cols)):
t=t+1
val = float(row[num_cols[i]]) + 10e-15
val = 0 if val < 0 else round(log(val),2)
num_hash = 'i' + str(num_cols[i]) + "_numlog_" + str(val)
x[t] = abs(hash(num_hash)) % D
for i in xrange(len(int_cols)):
t=t+1
val = float(row[int_cols[i]]) + 10e-15
val = 0 if val < 0 else round(log(val),2)
num_hash = 'i' + str(int_cols[i]) + "_numlog_" + str(val)
x[t] = abs(hash(num_hash)) % D
for i in xrange(len(zero_cols)):
t=t+1
val = float(row[zero_cols[i]]) + 10e-15
val = 0 if val < 0 else round(log(val),2)
num_hash = 'i' + str(zero_cols[i]) + "_numlog_" + str(val)
x[t] = abs(hash(num_hash)) % D
#+60+20+5
for i in xrange(len(num_cols)):
t=t+1
val = round(log(float(row[num_cols[i]]) ** 2 + 10e-15), 2)
num_hash = 'i' + str(num_cols[i]) + "_numlog_**2_" + str(val)
x[t] = abs(hash(num_hash)) % D
for i in xrange(len(int_cols)):
t=t+1
val = round(log(float(row[int_cols[i]]) ** 2 + 10e-15), 2)
num_hash = 'i' + str(int_cols[i]) + "_numlog_**2_" + str(val)
x[t] = abs(hash(num_hash)) % D
for i in xrange(len(zero_cols)):
t=t+1
val = round(log(float(row[zero_cols[i]]) ** 2 + 10e-15), 2)
num_hash = 'i' + str(zero_cols[i]) + "_numlog_**2_" + str(val)
x[t] = abs(hash(num_hash)) % D
# parse y, if provided
if label_path:
# use float() to prevent future type casting, [1:] to ignore id
y = [float(y) for y in label.readline().split(',')[1:]]
yield (ID, x, y) if label_path else (ID, x)
# B. Bounded logloss
# INPUT:
# p: our prediction
# y: real answer
# OUTPUT
# bounded logarithmic loss of p given y
def logloss(p, y):
p = max(min(p, 1. - 10e-15), 10e-15)
return -log(p) if y == 1. else -log(1. - p)
# C. Get probability estimation on x
# INPUT:
# x: features
# w: weights
# OUTPUT:
# probability of p(y = 1 | x; w)
def predict(x, w):
wTx = 0.
for i in x: # do wTx
wTx += w[i] * 1. # w[i] * x[i], but if i in x we got x[i] = 1.
return 1. / (1. + exp(-max(min(wTx, 20.), -20.))) # bounded sigmoid
# D. Update given model
# INPUT:
# alpha: learning rate
# w: weights
# n: sum of previous absolute gradients for a given feature
# this is used for adaptive learning rate
# x: feature, a list of indices
# p: prediction of our model
# y: answer
# MODIFIES:
# w: weights
# n: sum of past absolute gradients
def update(alpha, w, n, x, p, y):
for i in x:
# alpha / sqrt(n) is the adaptive learning rate
# (p - y) * x[i] is the current gradient
# note that in our case, if i in x then x[i] = 1.
n[i] += abs(p - y)
w[i] -= (p - y) * 1. * alpha / sqrt(n[i])
# training and testing #######################################################
start = datetime.now()
# a list for range(0, 33) - 13, no need to learn y14 since it is always 0
K = [k for k in range(33) if k not in [13,32]]
# initialize our model, all 32 of them, again ignoring y14
w = [[0.] * D if k not in [13,32] else None for k in range(33)]
n = [[0.] * D if k not in [13,32] else None for k in range(33)]
loss = 0.
loss_y14 = log(1. - 10**-15)
passNum = 0
lastLoss = 10.
thisLoss = 1.
while (lastLoss - thisLoss) > 0.000001 and passNum < 5:
lastLoss = thisLoss
passNum += 1
for ID, x, y in data(train, label):
ID = ID + 1700000*(passNum-1)
# get predictions and train on all labels
for k in K:
p = predict(x, w[k])
update(alpha, w[k], n[k], x, p, y[k])
loss += logloss(p, y[k]) # for progressive validation
loss += loss_y14 # the loss of y14, logloss is never zero
# print out progress, so that we know everything is working
if ID % 100000 == 0:
monitor.write('%s\tencountered: %d\tcurrent logloss: %f\n' % (
datetime.now(), ID, (loss/33.)/ID))
monitor.flush()
thisLoss = (loss/32)/ID
thisFile = './submission#'+str(solution)+'_'+str(passNum)+'.csv'
with open(thisFile, 'w') as outfile:
outfile.write('id_label,pred\n')
for ID, x in data(test):
predSum = 1.0
for k in K:
p = predict(x, w[k])
outfile.write('%s_y%d,%s\n' % (ID, k+1, str(p)))
predSum -= p
if k == 12:
outfile.write('%s_y14,0.0\n' % ID)
if k == 31:
p = max(0.01,predSum)
outfile.write('%s_y33,%s\n' % (ID, str(p)))
monitor.write('Done, elapsed time: %s\n' % str(datetime.now() - start))
monitor.close()