質問編集履歴
5
ｄ
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -66,6 +66,8 @@
|
|
|
66
66
|
|
|
67
67
|
```
|
|
68
68
|
|
|
69
|
+
全コード
|
|
70
|
+
|
|
69
71
|
# main module
|
|
70
72
|
|
|
71
73
|
import matplotlib.pyplot as plt
|
4
あ
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -14,13 +14,175 @@
|
|
|
14
14
|
|
|
15
15
|
|
|
16
16
|
|
|
17
|
-
その時の学習実行コードは以下の通りです。
|
|
17
|
+
その時の学習実行コードは以下の通りです。（学習実行コードは下のほうへ移動しました。）
|
|
18
18
|
|
|
19
19
|
|
|
20
20
|
|
|
21
21
|
この大量のメモリ消費の原因と、解決策をご教授いただけませんか。
|
|
22
22
|
|
|
23
|
+
追記事項
|
|
24
|
+
|
|
25
|
+
バージョン
|
|
26
|
+
|
|
23
|
-
```
|
|
27
|
+
```
|
|
28
|
+
|
|
29
|
+
Python 3.8.10
|
|
30
|
+
|
|
31
|
+
pandas 1.2.5
|
|
32
|
+
|
|
33
|
+
lightgbm 3.1.1
|
|
34
|
+
|
|
35
|
+
notebook 6.4.0 py38haa95532_0
|
|
36
|
+
|
|
37
|
+
numba 0.53.1 py38hf11a4ad_0
|
|
38
|
+
|
|
39
|
+
numpy 1.20.2 py38ha4e8547_0
|
|
40
|
+
|
|
41
|
+
numpy-base 1.20.2 py38hc2deb75_0
|
|
42
|
+
|
|
43
|
+
```
|
|
44
|
+
|
|
45
|
+
生のトレインデータ
|
|
46
|
+
|
|
47
|
+
|
|
48
|
+
|
|
49
|
+
|
|
50
|
+
|
|
51
|
+
加工後のトレインデータ
|
|
52
|
+
|
|
53
|
+
|
|
54
|
+
|
|
55
|
+
|
|
56
|
+
|
|
57
|
+
それぞれva_period = 1で実行時の
|
|
58
|
+
|
|
59
|
+
訓練データ、検証データ
|
|
60
|
+
|
|
61
|
+
訓練データのｙ、検証データのｙの大きさ
|
|
62
|
+
|
|
63
|
+
(254302, 10) (254302, 10)
|
|
64
|
+
|
|
65
|
+
(254302, 1) (254302, 1)
|
|
66
|
+
|
|
67
|
+
```
|
|
68
|
+
|
|
69
|
+
# main module
|
|
70
|
+
|
|
71
|
+
import matplotlib.pyplot as plt
|
|
72
|
+
|
|
73
|
+
import seaborn as sns
|
|
74
|
+
|
|
75
|
+
import pandas as pd
|
|
76
|
+
|
|
77
|
+
import pandas_profiling as pdp
|
|
78
|
+
|
|
79
|
+
import numpy as np
|
|
80
|
+
|
|
81
|
+
|
|
82
|
+
|
|
83
|
+
import lightgbm as lgb
|
|
84
|
+
|
|
85
|
+
from sklearn.metrics import log_loss
|
|
86
|
+
|
|
87
|
+
|
|
88
|
+
|
|
89
|
+
import datetime
|
|
90
|
+
|
|
91
|
+
import logging
|
|
92
|
+
|
|
93
|
+
import sys, os
|
|
94
|
+
|
|
95
|
+
sys.path.append('../src/')
|
|
96
|
+
|
|
97
|
+
|
|
98
|
+
|
|
99
|
+
import eda
|
|
100
|
+
|
|
101
|
+
import maprepro as mpre
|
|
102
|
+
|
|
103
|
+
# import config
|
|
104
|
+
|
|
105
|
+
# from utils import setup_logger, ModelFactory
|
|
106
|
+
|
|
107
|
+
path = '../../input'
|
|
108
|
+
|
|
109
|
+
sample = pd.read_csv(f'{path}/sample_submission.csv')
|
|
110
|
+
|
|
111
|
+
store = pd.read_csv(f'{path}/store.csv')
|
|
112
|
+
|
|
113
|
+
test = pd.read_csv(f'{path}/test.csv')
|
|
114
|
+
|
|
115
|
+
train = pd.read_csv(f'{path}/train.csv')
|
|
116
|
+
|
|
117
|
+
|
|
118
|
+
|
|
119
|
+
def mk_ymd(df):
|
|
120
|
+
|
|
121
|
+
df['year'] = df.Date.apply(lambda x: x.split('-')[0]).astype(np.int16)
|
|
122
|
+
|
|
123
|
+
df['month'] = df.Date.apply(lambda x: x.split('-')[1]).astype(np.int16)
|
|
124
|
+
|
|
125
|
+
df['day'] = df.Date.apply(lambda x: x.split('-')[2]).astype(np.int16)
|
|
126
|
+
|
|
127
|
+
df = df.sort_values('Date')
|
|
128
|
+
|
|
129
|
+
return df
|
|
130
|
+
|
|
131
|
+
train = mk_ymd(train)
|
|
132
|
+
|
|
133
|
+
test = mk_ymd(test)
|
|
134
|
+
|
|
135
|
+
|
|
136
|
+
|
|
137
|
+
|
|
138
|
+
|
|
139
|
+
# 時系列データであり、時間に沿って変数periodを設定したとする
|
|
140
|
+
|
|
141
|
+
train['period'] = np.arange(0, len(train)) // (len(train) // 4)
|
|
142
|
+
|
|
143
|
+
train['period'] = np.clip(train['period'], 0, 3)
|
|
144
|
+
|
|
145
|
+
test['period'] = 4
|
|
146
|
+
|
|
147
|
+
|
|
148
|
+
|
|
149
|
+
train['StateHoliday'] = train.StateHoliday.astype('category')
|
|
150
|
+
|
|
151
|
+
|
|
152
|
+
|
|
153
|
+
target = ['Sales']
|
|
154
|
+
|
|
155
|
+
notuse = ['Id','Date','Open']
|
|
156
|
+
|
|
157
|
+
use = ['Store','DayOfWeek','Open','Promo','StateHoliday','SchoolHoliday','year','month','day','period']
|
|
158
|
+
|
|
159
|
+
|
|
160
|
+
|
|
161
|
+
train_y = train[target]
|
|
162
|
+
|
|
163
|
+
train_x = train[use]
|
|
164
|
+
|
|
165
|
+
test_x = test[use]
|
|
166
|
+
|
|
167
|
+
|
|
168
|
+
|
|
169
|
+
|
|
170
|
+
|
|
171
|
+
import warnings
|
|
172
|
+
|
|
173
|
+
warnings.simplefilter('ignore')
|
|
174
|
+
|
|
175
|
+
train_x = eda.reduce_mem_usage(train_x)
|
|
176
|
+
|
|
177
|
+
>>>start size(BEFORE): 57.24 Mb
|
|
178
|
+
|
|
179
|
+
>>>Mem. usage decreased to 19.40 Mb (AFTER:66.1% reduction)
|
|
180
|
+
|
|
181
|
+
|
|
182
|
+
|
|
183
|
+
import gc
|
|
184
|
+
|
|
185
|
+
gc.collect()
|
|
24
186
|
|
|
25
187
|
va_period_list = [1, 2, 3]
|
|
26
188
|
|
|
@@ -34,6 +196,10 @@
|
|
|
34
196
|
|
|
35
197
|
tr_y, va_y = train_y[is_tr], train_y[is_va]
|
|
36
198
|
|
|
199
|
+
print(tr_x.shape, va_x.shape)
|
|
200
|
+
|
|
201
|
+
print(tr_y.shape, va_y.shape)
|
|
202
|
+
|
|
37
203
|
|
|
38
204
|
|
|
39
205
|
lgb_train = lgb.Dataset(tr_x, tr_y)
|
|
@@ -48,7 +214,7 @@
|
|
|
48
214
|
|
|
49
215
|
'seed': 71,
|
|
50
216
|
|
|
51
|
-
'verbose':
|
|
217
|
+
'verbose': 1,
|
|
52
218
|
|
|
53
219
|
'metrics': 'binary_logloss',
|
|
54
220
|
|
|
@@ -56,7 +222,7 @@
|
|
|
56
222
|
|
|
57
223
|
}
|
|
58
224
|
|
|
59
|
-
num_round = 10
|
|
225
|
+
num_round = 10
|
|
60
226
|
|
|
61
227
|
|
|
62
228
|
|
|
@@ -74,8 +240,6 @@
|
|
|
74
240
|
|
|
75
241
|
valid_names=['train', 'valid'], valid_sets=[lgb_train, lgb_eval],
|
|
76
242
|
|
|
77
|
-
|
|
78
|
-
|
|
79
243
|
)
|
|
80
244
|
|
|
81
245
|
|
|
@@ -94,242 +258,6 @@
|
|
|
94
258
|
|
|
95
259
|
pred = model.predict(test_x)
|
|
96
260
|
|
|
261
|
+
|
|
262
|
+
|
|
97
|
-
```
|
|
263
|
+
```
|
|
98
|
-
|
|
99
|
-
|
|
100
|
-
|
|
101
|
-
追記事項
|
|
102
|
-
|
|
103
|
-
バージョン
|
|
104
|
-
|
|
105
|
-
```
|
|
106
|
-
|
|
107
|
-
Python 3.8.10
|
|
108
|
-
|
|
109
|
-
pandas 1.2.5
|
|
110
|
-
|
|
111
|
-
lightgbm 3.1.1
|
|
112
|
-
|
|
113
|
-
notebook 6.4.0 py38haa95532_0
|
|
114
|
-
|
|
115
|
-
numba 0.53.1 py38hf11a4ad_0
|
|
116
|
-
|
|
117
|
-
numpy 1.20.2 py38ha4e8547_0
|
|
118
|
-
|
|
119
|
-
numpy-base 1.20.2 py38hc2deb75_0
|
|
120
|
-
|
|
121
|
-
```
|
|
122
|
-
|
|
123
|
-
生のトレインデータ
|
|
124
|
-
|
|
125
|
-
|
|
126
|
-
|
|
127
|
-
|
|
128
|
-
|
|
129
|
-
それぞれva_period = 1で実行時の
|
|
130
|
-
|
|
131
|
-
訓練データ、検証データ
|
|
132
|
-
|
|
133
|
-
訓練データのｙ、検証データのｙの大きさ
|
|
134
|
-
|
|
135
|
-
(254302, 10) (254302, 10)
|
|
136
|
-
|
|
137
|
-
(254302, 1) (254302, 1)
|
|
138
|
-
|
|
139
|
-
```
|
|
140
|
-
|
|
141
|
-
# main module
|
|
142
|
-
|
|
143
|
-
import matplotlib.pyplot as plt
|
|
144
|
-
|
|
145
|
-
import seaborn as sns
|
|
146
|
-
|
|
147
|
-
import pandas as pd
|
|
148
|
-
|
|
149
|
-
import pandas_profiling as pdp
|
|
150
|
-
|
|
151
|
-
import numpy as np
|
|
152
|
-
|
|
153
|
-
|
|
154
|
-
|
|
155
|
-
import lightgbm as lgb
|
|
156
|
-
|
|
157
|
-
from sklearn.metrics import log_loss
|
|
158
|
-
|
|
159
|
-
|
|
160
|
-
|
|
161
|
-
import datetime
|
|
162
|
-
|
|
163
|
-
import logging
|
|
164
|
-
|
|
165
|
-
import sys, os
|
|
166
|
-
|
|
167
|
-
sys.path.append('../src/')
|
|
168
|
-
|
|
169
|
-
|
|
170
|
-
|
|
171
|
-
import eda
|
|
172
|
-
|
|
173
|
-
import maprepro as mpre
|
|
174
|
-
|
|
175
|
-
# import config
|
|
176
|
-
|
|
177
|
-
# from utils import setup_logger, ModelFactory
|
|
178
|
-
|
|
179
|
-
path = '../../input'
|
|
180
|
-
|
|
181
|
-
sample = pd.read_csv(f'{path}/sample_submission.csv')
|
|
182
|
-
|
|
183
|
-
store = pd.read_csv(f'{path}/store.csv')
|
|
184
|
-
|
|
185
|
-
test = pd.read_csv(f'{path}/test.csv')
|
|
186
|
-
|
|
187
|
-
train = pd.read_csv(f'{path}/train.csv')
|
|
188
|
-
|
|
189
|
-
|
|
190
|
-
|
|
191
|
-
def mk_ymd(df):
|
|
192
|
-
|
|
193
|
-
df['year'] = df.Date.apply(lambda x: x.split('-')[0]).astype(np.int16)
|
|
194
|
-
|
|
195
|
-
df['month'] = df.Date.apply(lambda x: x.split('-')[1]).astype(np.int16)
|
|
196
|
-
|
|
197
|
-
df['day'] = df.Date.apply(lambda x: x.split('-')[2]).astype(np.int16)
|
|
198
|
-
|
|
199
|
-
df = df.sort_values('Date')
|
|
200
|
-
|
|
201
|
-
return df
|
|
202
|
-
|
|
203
|
-
train = mk_ymd(train)
|
|
204
|
-
|
|
205
|
-
test = mk_ymd(test)
|
|
206
|
-
|
|
207
|
-
|
|
208
|
-
|
|
209
|
-
|
|
210
|
-
|
|
211
|
-
# 時系列データであり、時間に沿って変数periodを設定したとする
|
|
212
|
-
|
|
213
|
-
train['period'] = np.arange(0, len(train)) // (len(train) // 4)
|
|
214
|
-
|
|
215
|
-
train['period'] = np.clip(train['period'], 0, 3)
|
|
216
|
-
|
|
217
|
-
test['period'] = 4
|
|
218
|
-
|
|
219
|
-
|
|
220
|
-
|
|
221
|
-
train['StateHoliday'] = train.StateHoliday.astype('category')
|
|
222
|
-
|
|
223
|
-
|
|
224
|
-
|
|
225
|
-
target = ['Sales']
|
|
226
|
-
|
|
227
|
-
notuse = ['Id','Date','Open']
|
|
228
|
-
|
|
229
|
-
use = ['Store','DayOfWeek','Open','Promo','StateHoliday','SchoolHoliday','year','month','day','period']
|
|
230
|
-
|
|
231
|
-
|
|
232
|
-
|
|
233
|
-
train_y = train[target]
|
|
234
|
-
|
|
235
|
-
train_x = train[use]
|
|
236
|
-
|
|
237
|
-
test_x = test[use]
|
|
238
|
-
|
|
239
|
-
|
|
240
|
-
|
|
241
|
-
|
|
242
|
-
|
|
243
|
-
import warnings
|
|
244
|
-
|
|
245
|
-
warnings.simplefilter('ignore')
|
|
246
|
-
|
|
247
|
-
train_x = eda.reduce_mem_usage(train_x)
|
|
248
|
-
|
|
249
|
-
>>>start size(BEFORE): 57.24 Mb
|
|
250
|
-
|
|
251
|
-
>>>Mem. usage decreased to 19.40 Mb (AFTER:66.1% reduction)
|
|
252
|
-
|
|
253
|
-
|
|
254
|
-
|
|
255
|
-
import gc
|
|
256
|
-
|
|
257
|
-
gc.collect()
|
|
258
|
-
|
|
259
|
-
va_period_list = [1, 2, 3]
|
|
260
|
-
|
|
261
|
-
for va_period in va_period_list:
|
|
262
|
-
|
|
263
|
-
is_tr = train_x['period'] < va_period
|
|
264
|
-
|
|
265
|
-
is_va = train_x['period'] == va_period
|
|
266
|
-
|
|
267
|
-
tr_x, va_x = train_x[is_tr], train_x[is_va]
|
|
268
|
-
|
|
269
|
-
tr_y, va_y = train_y[is_tr], train_y[is_va]
|
|
270
|
-
|
|
271
|
-
print(tr_x.shape, va_x.shape)
|
|
272
|
-
|
|
273
|
-
print(tr_y.shape, va_y.shape)
|
|
274
|
-
|
|
275
|
-
|
|
276
|
-
|
|
277
|
-
lgb_train = lgb.Dataset(tr_x, tr_y)
|
|
278
|
-
|
|
279
|
-
lgb_eval = lgb.Dataset(va_x, va_y)
|
|
280
|
-
|
|
281
|
-
|
|
282
|
-
|
|
283
|
-
# ハイパーパラメータの設定
|
|
284
|
-
|
|
285
|
-
params = {'objective': 'binary',
|
|
286
|
-
|
|
287
|
-
'seed': 71,
|
|
288
|
-
|
|
289
|
-
'verbose': 1,
|
|
290
|
-
|
|
291
|
-
'metrics': 'binary_logloss',
|
|
292
|
-
|
|
293
|
-
'force_col_wise':'true' # メモリが足りないから
|
|
294
|
-
|
|
295
|
-
}
|
|
296
|
-
|
|
297
|
-
num_round = 10
|
|
298
|
-
|
|
299
|
-
|
|
300
|
-
|
|
301
|
-
# 学習の実行
|
|
302
|
-
|
|
303
|
-
# カテゴリ変数をパラメータで指定している
|
|
304
|
-
|
|
305
|
-
# バリデーションデータもモデルに渡し、学習の進行とともにスコアがどう変わるかモニタリングする
|
|
306
|
-
|
|
307
|
-
categorical_features = ['StateHoliday']
|
|
308
|
-
|
|
309
|
-
model = lgb.train(params, lgb_train, num_boost_round=num_round,
|
|
310
|
-
|
|
311
|
-
categorical_feature=categorical_features,
|
|
312
|
-
|
|
313
|
-
valid_names=['train', 'valid'], valid_sets=[lgb_train, lgb_eval],
|
|
314
|
-
|
|
315
|
-
)
|
|
316
|
-
|
|
317
|
-
|
|
318
|
-
|
|
319
|
-
# バリデーションデータでのスコアの確認
|
|
320
|
-
|
|
321
|
-
va_pred = model.predict(va_x)
|
|
322
|
-
|
|
323
|
-
score = log_loss(va_y, va_pred)
|
|
324
|
-
|
|
325
|
-
print(f'logloss: {score:.4f}')
|
|
326
|
-
|
|
327
|
-
|
|
328
|
-
|
|
329
|
-
# 予測
|
|
330
|
-
|
|
331
|
-
pred = model.predict(test_x)
|
|
332
|
-
|
|
333
|
-
|
|
334
|
-
|
|
335
|
-
```
|
3
あ
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -126,9 +126,15 @@
|
|
|
126
126
|
|
|
127
127
|
|
|
128
128
|
|
|
129
|
-
|
|
129
|
+
それぞれva_period = 1で実行時の
|
|
130
|
+
|
|
130
|
-
|
|
131
|
+
訓練データ、検証データ
|
|
132
|
+
|
|
131
|
-
|
|
133
|
+
訓練データのｙ、検証データのｙの大きさ
|
|
134
|
+
|
|
135
|
+
(254302, 10) (254302, 10)
|
|
136
|
+
|
|
137
|
+
(254302, 1) (254302, 1)
|
|
132
138
|
|
|
133
139
|
```
|
|
134
140
|
|
2
あ
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -120,6 +120,14 @@
|
|
|
120
120
|
|
|
121
121
|
```
|
|
122
122
|
|
|
123
|
+
生のトレインデータ
|
|
124
|
+
|
|
125
|
+
|
|
126
|
+
|
|
127
|
+
|
|
128
|
+
|
|
129
|
+
|
|
130
|
+
|
|
123
131
|
|
|
124
132
|
|
|
125
133
|
```
|
|
@@ -318,12 +326,4 @@
|
|
|
318
326
|
|
|
319
327
|
|
|
320
328
|
|
|
321
|
-
|
|
322
|
-
|
|
323
|
-
|
|
324
|
-
|
|
325
|
-
|
|
326
|
-
|
|
327
|
-
|
|
328
|
-
|
|
329
|
-
```
|
|
329
|
+
```
|
1
s
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -95,3 +95,235 @@
|
|
|
95
95
|
pred = model.predict(test_x)
|
|
96
96
|
|
|
97
97
|
```
|
|
98
|
+
|
|
99
|
+
|
|
100
|
+
|
|
101
|
+
追記事項
|
|
102
|
+
|
|
103
|
+
バージョン
|
|
104
|
+
|
|
105
|
+
```
|
|
106
|
+
|
|
107
|
+
Python 3.8.10
|
|
108
|
+
|
|
109
|
+
pandas 1.2.5
|
|
110
|
+
|
|
111
|
+
lightgbm 3.1.1
|
|
112
|
+
|
|
113
|
+
notebook 6.4.0 py38haa95532_0
|
|
114
|
+
|
|
115
|
+
numba 0.53.1 py38hf11a4ad_0
|
|
116
|
+
|
|
117
|
+
numpy 1.20.2 py38ha4e8547_0
|
|
118
|
+
|
|
119
|
+
numpy-base 1.20.2 py38hc2deb75_0
|
|
120
|
+
|
|
121
|
+
```
|
|
122
|
+
|
|
123
|
+
|
|
124
|
+
|
|
125
|
+
```
|
|
126
|
+
|
|
127
|
+
# main module
|
|
128
|
+
|
|
129
|
+
import matplotlib.pyplot as plt
|
|
130
|
+
|
|
131
|
+
import seaborn as sns
|
|
132
|
+
|
|
133
|
+
import pandas as pd
|
|
134
|
+
|
|
135
|
+
import pandas_profiling as pdp
|
|
136
|
+
|
|
137
|
+
import numpy as np
|
|
138
|
+
|
|
139
|
+
|
|
140
|
+
|
|
141
|
+
import lightgbm as lgb
|
|
142
|
+
|
|
143
|
+
from sklearn.metrics import log_loss
|
|
144
|
+
|
|
145
|
+
|
|
146
|
+
|
|
147
|
+
import datetime
|
|
148
|
+
|
|
149
|
+
import logging
|
|
150
|
+
|
|
151
|
+
import sys, os
|
|
152
|
+
|
|
153
|
+
sys.path.append('../src/')
|
|
154
|
+
|
|
155
|
+
|
|
156
|
+
|
|
157
|
+
import eda
|
|
158
|
+
|
|
159
|
+
import maprepro as mpre
|
|
160
|
+
|
|
161
|
+
# import config
|
|
162
|
+
|
|
163
|
+
# from utils import setup_logger, ModelFactory
|
|
164
|
+
|
|
165
|
+
path = '../../input'
|
|
166
|
+
|
|
167
|
+
sample = pd.read_csv(f'{path}/sample_submission.csv')
|
|
168
|
+
|
|
169
|
+
store = pd.read_csv(f'{path}/store.csv')
|
|
170
|
+
|
|
171
|
+
test = pd.read_csv(f'{path}/test.csv')
|
|
172
|
+
|
|
173
|
+
train = pd.read_csv(f'{path}/train.csv')
|
|
174
|
+
|
|
175
|
+
|
|
176
|
+
|
|
177
|
+
def mk_ymd(df):
|
|
178
|
+
|
|
179
|
+
df['year'] = df.Date.apply(lambda x: x.split('-')[0]).astype(np.int16)
|
|
180
|
+
|
|
181
|
+
df['month'] = df.Date.apply(lambda x: x.split('-')[1]).astype(np.int16)
|
|
182
|
+
|
|
183
|
+
df['day'] = df.Date.apply(lambda x: x.split('-')[2]).astype(np.int16)
|
|
184
|
+
|
|
185
|
+
df = df.sort_values('Date')
|
|
186
|
+
|
|
187
|
+
return df
|
|
188
|
+
|
|
189
|
+
train = mk_ymd(train)
|
|
190
|
+
|
|
191
|
+
test = mk_ymd(test)
|
|
192
|
+
|
|
193
|
+
|
|
194
|
+
|
|
195
|
+
|
|
196
|
+
|
|
197
|
+
# 時系列データであり、時間に沿って変数periodを設定したとする
|
|
198
|
+
|
|
199
|
+
train['period'] = np.arange(0, len(train)) // (len(train) // 4)
|
|
200
|
+
|
|
201
|
+
train['period'] = np.clip(train['period'], 0, 3)
|
|
202
|
+
|
|
203
|
+
test['period'] = 4
|
|
204
|
+
|
|
205
|
+
|
|
206
|
+
|
|
207
|
+
train['StateHoliday'] = train.StateHoliday.astype('category')
|
|
208
|
+
|
|
209
|
+
|
|
210
|
+
|
|
211
|
+
target = ['Sales']
|
|
212
|
+
|
|
213
|
+
notuse = ['Id','Date','Open']
|
|
214
|
+
|
|
215
|
+
use = ['Store','DayOfWeek','Open','Promo','StateHoliday','SchoolHoliday','year','month','day','period']
|
|
216
|
+
|
|
217
|
+
|
|
218
|
+
|
|
219
|
+
train_y = train[target]
|
|
220
|
+
|
|
221
|
+
train_x = train[use]
|
|
222
|
+
|
|
223
|
+
test_x = test[use]
|
|
224
|
+
|
|
225
|
+
|
|
226
|
+
|
|
227
|
+
|
|
228
|
+
|
|
229
|
+
import warnings
|
|
230
|
+
|
|
231
|
+
warnings.simplefilter('ignore')
|
|
232
|
+
|
|
233
|
+
train_x = eda.reduce_mem_usage(train_x)
|
|
234
|
+
|
|
235
|
+
>>>start size(BEFORE): 57.24 Mb
|
|
236
|
+
|
|
237
|
+
>>>Mem. usage decreased to 19.40 Mb (AFTER:66.1% reduction)
|
|
238
|
+
|
|
239
|
+
|
|
240
|
+
|
|
241
|
+
import gc
|
|
242
|
+
|
|
243
|
+
gc.collect()
|
|
244
|
+
|
|
245
|
+
va_period_list = [1, 2, 3]
|
|
246
|
+
|
|
247
|
+
for va_period in va_period_list:
|
|
248
|
+
|
|
249
|
+
is_tr = train_x['period'] < va_period
|
|
250
|
+
|
|
251
|
+
is_va = train_x['period'] == va_period
|
|
252
|
+
|
|
253
|
+
tr_x, va_x = train_x[is_tr], train_x[is_va]
|
|
254
|
+
|
|
255
|
+
tr_y, va_y = train_y[is_tr], train_y[is_va]
|
|
256
|
+
|
|
257
|
+
print(tr_x.shape, va_x.shape)
|
|
258
|
+
|
|
259
|
+
print(tr_y.shape, va_y.shape)
|
|
260
|
+
|
|
261
|
+
|
|
262
|
+
|
|
263
|
+
lgb_train = lgb.Dataset(tr_x, tr_y)
|
|
264
|
+
|
|
265
|
+
lgb_eval = lgb.Dataset(va_x, va_y)
|
|
266
|
+
|
|
267
|
+
|
|
268
|
+
|
|
269
|
+
# ハイパーパラメータの設定
|
|
270
|
+
|
|
271
|
+
params = {'objective': 'binary',
|
|
272
|
+
|
|
273
|
+
'seed': 71,
|
|
274
|
+
|
|
275
|
+
'verbose': 1,
|
|
276
|
+
|
|
277
|
+
'metrics': 'binary_logloss',
|
|
278
|
+
|
|
279
|
+
'force_col_wise':'true' # メモリが足りないから
|
|
280
|
+
|
|
281
|
+
}
|
|
282
|
+
|
|
283
|
+
num_round = 10
|
|
284
|
+
|
|
285
|
+
|
|
286
|
+
|
|
287
|
+
# 学習の実行
|
|
288
|
+
|
|
289
|
+
# カテゴリ変数をパラメータで指定している
|
|
290
|
+
|
|
291
|
+
# バリデーションデータもモデルに渡し、学習の進行とともにスコアがどう変わるかモニタリングする
|
|
292
|
+
|
|
293
|
+
categorical_features = ['StateHoliday']
|
|
294
|
+
|
|
295
|
+
model = lgb.train(params, lgb_train, num_boost_round=num_round,
|
|
296
|
+
|
|
297
|
+
categorical_feature=categorical_features,
|
|
298
|
+
|
|
299
|
+
valid_names=['train', 'valid'], valid_sets=[lgb_train, lgb_eval],
|
|
300
|
+
|
|
301
|
+
)
|
|
302
|
+
|
|
303
|
+
|
|
304
|
+
|
|
305
|
+
# バリデーションデータでのスコアの確認
|
|
306
|
+
|
|
307
|
+
va_pred = model.predict(va_x)
|
|
308
|
+
|
|
309
|
+
score = log_loss(va_y, va_pred)
|
|
310
|
+
|
|
311
|
+
print(f'logloss: {score:.4f}')
|
|
312
|
+
|
|
313
|
+
|
|
314
|
+
|
|
315
|
+
# 予測
|
|
316
|
+
|
|
317
|
+
pred = model.predict(test_x)
|
|
318
|
+
|
|
319
|
+
|
|
320
|
+
|
|
321
|
+
|
|
322
|
+
|
|
323
|
+
|
|
324
|
+
|
|
325
|
+
|
|
326
|
+
|
|
327
|
+
|
|
328
|
+
|
|
329
|
+
```
|