-
Notifications
You must be signed in to change notification settings - Fork 57
/
990_2nd_place_solution_golf.py
272 lines (199 loc) · 7.6 KB
/
990_2nd_place_solution_golf.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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Thu Apr 11 13:36:18 2019
@author: kazuki.onodera
└── input
├── sample_submission.csv.zip
├── test.csv.zip
└── train.csv.zip
"""
import numpy as np
import pandas as pd
import gc, os
import lightgbm as lgb
from sklearn.metrics import roc_auc_score
from sklearn.model_selection import StratifiedKFold
from sklearn.preprocessing import StandardScaler
from multiprocessing import cpu_count
from tqdm import tqdm
# ===== fake samples =====
te_ = pd.read_csv('../input/test.csv.zip').drop(['ID_code'], axis=1).values
unique_samples = []
unique_count = np.zeros_like(te_)
for feature in tqdm(range(te_.shape[1])):
_, index_, count_ = np.unique(te_[:, feature], return_counts=True, return_index=True)
unique_count[index_[count_ == 1], feature] += 1
# Samples which have unique values are real the others are fake
real_samples_indexes = np.argwhere(np.sum(unique_count, axis=1) > 0)[:, 0]
synthetic_samples_indexes = np.argwhere(np.sum(unique_count, axis=1) == 0)[:, 0]
# =============================================================================
# setting
# =============================================================================
# parameters
params = {
'bagging_freq': 5,
'bagging_fraction': 1.0,
'boost_from_average':'false',
'boost': 'gbdt',
'feature_fraction': 1.0,
'learning_rate': 0.005,
'max_depth': -1,
'metric':'binary_logloss',
'min_data_in_leaf': 30,
'min_sum_hessian_in_leaf': 10.0,
'num_leaves': 64,
'num_threads': cpu_count(),
'tree_learner': 'serial',
'objective': 'binary',
'verbosity': -1
}
NFOLD = 10
NROUND = 1600
SEED = np.random.randint(99999)
np.random.seed(SEED)
SUBMIT_FILE_PATH = f'../output/2nd-place-solution.csv.gz'
# =============================================================================
# drop vars
# =============================================================================
drop_vars = [7,
10,
17,
27,
29,
30,
38,
41,
46,
96,
100,
103,
126,
158,
185]
var_len = 200 - len(drop_vars)
# =============================================================================
# load
# =============================================================================
train = pd.read_csv("../input/train.csv.zip")
test = pd.read_csv("../input/test.csv.zip").drop(synthetic_samples_indexes)
X_train = train.iloc[:, 2:].values
y_train = train.target.values
X_test = test.iloc[:, 1:].values
X = np.concatenate([X_train, X_test], axis=0)
del X_train, X_test; gc.collect()
reverse_list = [0, 1, 2, 3, 4, 5, 6, 7, 8, 11, 15, 16, 18, 19, 22, 24, 25, 26,
27, 29, 32, 35, 37, 40, 41, 47, 48, 49, 51, 52, 53, 55, 60, 61,
62, 65, 66, 67, 69, 70, 71, 74, 78, 79, 82, 84, 89, 90, 91, 94,
95, 96, 97, 99, 103, 105, 106, 110, 111, 112, 118, 119, 125, 128,
130, 133, 134, 135, 137, 138, 140, 144, 145, 147, 151, 155, 157,
159, 161, 162, 163, 164, 167, 168, 170, 171, 173, 175, 176, 179,
180, 181, 184, 185, 187, 189, 190, 191, 195, 196, 199,
]
for j in reverse_list:
X[:, j] *= -1
# drop
X = np.delete(X, drop_vars, 1)
# scaling
scaler = StandardScaler()
X = scaler.fit_transform(X)
# count encoding
X_cnt = np.zeros((len(X), var_len * 4))
for j in tqdm(range(var_len)):
for i in range(1, 4):
x = np.round(X[:, j], i+1)
dic = pd.value_counts(x).to_dict()
X_cnt[:, i+j*4] = pd.Series(x).map(dic)
x = X[:, j]
dic = pd.value_counts(x).to_dict()
X_cnt[:, j*4] = pd.Series(x).map(dic)
# raw + count feature
X_raw = X.copy() # rename for readable
del X; gc.collect()
X = np.zeros((len(X_raw), var_len * 5))
for j in tqdm(range(var_len)):
X[:, 5*j+1:5*j+5] = X_cnt[:, 4*j:4*j+4]
X[:, 5*j] = X_raw[:, j]
# treat each var as same
X_train_concat = np.concatenate([
np.concatenate([
X[:200000, 5*cnum:5*cnum+5],
np.ones((len(y_train), 1)).astype("int")*cnum
], axis=1) for cnum in range(var_len)], axis=0)
y_train_concat = np.concatenate([y_train for cnum in range(var_len)], axis=0)
# =============================================================================
# stratified
# =============================================================================
train_group = np.arange(len(X_train_concat))%200000
id_y = pd.DataFrame(zip(train_group, y_train_concat),
columns=['id', 'y'])
id_y_uq = id_y.drop_duplicates('id').reset_index(drop=True)
def stratified(nfold=5):
id_y_uq0 = id_y_uq[id_y_uq.y==0].sample(frac=1)
id_y_uq1 = id_y_uq[id_y_uq.y==1].sample(frac=1)
id_y_uq0['g'] = [i%nfold for i in range(len(id_y_uq0))]
id_y_uq1['g'] = [i%nfold for i in range(len(id_y_uq1))]
id_y_uq_ = pd.concat([id_y_uq0, id_y_uq1])
id_y_ = pd.merge(id_y[['id']], id_y_uq_, how='left', on='id')
train_idx_list = []
valid_idx_list = []
for i in range(nfold):
train_idx = id_y_[id_y_.g!=i].index
train_idx_list.append(train_idx)
valid_idx = id_y_[id_y_.g==i].index
valid_idx_list.append(valid_idx)
return train_idx_list, valid_idx_list
train_idx_list, valid_idx_list = stratified(NFOLD)
# =============================================================================
# train
# =============================================================================
models = []
oof = np.zeros(len(id_y))
p_test_all = np.zeros((100000, var_len, NFOLD))
id_y['var'] = np.concatenate([np.ones(200000)*i for i in range(var_len)])
for i in range(NFOLD):
print(f'building {i}...')
train_idx = train_idx_list[i]
valid_idx = valid_idx_list[i]
# train
X_train_cv = X_train_concat[train_idx]
y_train_cv = y_train_concat[train_idx]
# valid
X_valid = X_train_concat[valid_idx]
# test
X_test = np.concatenate([
np.concatenate([
X[200000:, 5*cnum:5*cnum+5],
np.ones((100000, 1)).astype("int")*cnum
], axis=1) for cnum in range(var_len)], axis=0
)
dtrain = lgb.Dataset(
X_train_cv, y_train_cv,
feature_name=['value', 'count_org', 'count_2', 'count_3', 'count_4', 'varnum'],
categorical_feature=['varnum'], free_raw_data=False
)
model = lgb.train(params, train_set=dtrain, num_boost_round=NROUND, verbose_eval=100)
l = valid_idx.shape[0]
p_valid = model.predict(X_valid)
p_test = model.predict(X_test)
for j in range(var_len):
oof[valid_idx] = p_valid
p_test_all[:, j, i] = p_test[j*100000:(j+1)*100000]
models.append(model)
# =============================================================================
# test
# =============================================================================
id_y['pred'] = oof
oof = pd.pivot_table(id_y, index='id', columns='var', values='pred').values
p_test_mean = p_test_all.mean(axis=2)
p_test_odds = np.ones(100000) * 1 / 9
for j in range(var_len):
if roc_auc_score(y_train, oof[:, j]) >= 0.500:
p_test_odds *= (9 * p_test_mean[:, j] / (1 - p_test_mean[:, j]))
p_test_odds = p_test_odds / (1 + p_test_odds)
sub1 = pd.read_csv("../input/sample_submission.csv.zip")
sub2 = pd.DataFrame({"ID_code":test.ID_code.values , "target":p_test_odds})
sub = pd.merge(sub1[["ID_code"]], sub2, how="left").fillna(0)
# save
sub.to_csv(SUBMIT_FILE_PATH, index=False, compression='gzip')
#==============================================================================