forked from nlceyes/TianChi-JinNan
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Run-Playoffs.py
329 lines (319 loc) · 14.3 KB
/
Run-Playoffs.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
# -*- coding: utf-8 -*-
# Intial Run @20190218, Parameter Tuned Run @20190219
# Rev.1 @20180218: Categorical Features with pd.get_dummies - Seems Useless?
# Rev.2 @20180218: Adding Features
# Rev.3 @20180218: Adding Rules (B14=385,0.878)
# Rev.4 @20180218: Adding Features of B14 Statistical Info
# Rev.5 @20180218: Adding Stacking Method - Seems Useless? Using Avg Method
import time
from contextlib import contextmanager
import numpy as np
import pandas as pd
# from pandas.core.common import SettingWithCopyWarning
import lightgbm as lgb
import xgboost as xgb
from sklearn.metrics import mean_squared_error
from sklearn.linear_model import BayesianRidge, LinearRegression
# import warnings
# warnings.simplefilter(action='ignore', category=SettingWithCopyWarning)
# warnings.simplefilter(action='ignore', category=FutureWarning)
@contextmanager
def timer(title):
t0 = time.time()
yield
print('{} - Done in {:.1f}s'.format(title, time.time()-t0))
print('------------------------------------')
# Time Feature Processing
def get_time(t):
if pd.isnull(t):
pass
else:
# Get Valid Time and Split into Hrs and Mins
if ';' in t:
temp = t.split(';')
elif '::' in t:
temp = t.split('::')
elif '"' in t:
temp = t.split('"')
elif ';' in t:
temp = t.split(';')
elif ':' in t:
temp = t.split(':')
else:
temp = t.split(':')
# print(f'New Separate Character Found! - {t}')
# print(col)
# Hrs and Mins to Number
if len(temp) > 1:
hours = temp[0]
if len(temp[1]) == 0:
mins = 0
else:
mins = temp[1] if len(temp[1])<3 else temp[1][:2]
else:
hours = temp[0][:2]
mins = temp[0][2:]
time_int = int(hours) + int(mins)/60
return time_int
def get_period(t):
if pd.isnull(t):
pass
else:
start, end = t.split('-')
start = get_time(start)
end = get_time(end)
if end >= start:
time_period = end - start
else:
time_period = 24 + end - start
return time_period
# Loading and Preprocessing Dataset
def train_test():
# Loading Dataset and Target
df_submit = pd.read_csv('./data/FuSai.csv', encoding='gb18030')
df_opt = pd.read_csv('./data/optimize.csv', encoding='gb18030')
df_submit = df_submit.append(df_opt)
df_submit['收率'] = 0
df_train = pd.read_csv('./data/jinnan_round1_train_20181227.csv', encoding='gb18030')
df_A = pd.read_csv('./data/jinnan_round1_testA_20181227.csv', encoding='gb18030')
df_A_ans = pd.read_csv('./data/jinnan_round1_ansA_20190125.csv', header=None)
df_A['收率'] = df_A_ans[1]
df_B = pd.read_csv('./data/jinnan_round1_testB_20190121.csv', encoding='gb18030')
df_B_ans = pd.read_csv('./data/jinnan_round1_ansB_20190125.csv', header=None)
df_B['收率'] = df_B_ans[1]
df_C = pd.read_csv('./data/jinnan_round1_test_20190201.csv', encoding='gb18030')
df_C_ans = pd.read_csv('./data/jinnan_round1_ans_20190201.csv', header=None)
df_C['收率'] = df_C_ans[1]
print('Original:\nTrain samples: {}, Test samples: {},{},{}'.format(len(df_train), len(df_A), len(df_B), len(df_C)))
# Anomaly Processing
df_train.loc[df_train['A5']=='1900/1/29 0:00', 'A5'] = '0:00' # sample_223
df_train.loc[df_train['A5']=='1900/1/21 0:00', 'A5'] = '0:00' # sample_1023,1027
df_A.loc[df_A['A5']=='1900/1/22 0:00', 'A5'] = '0:00' # sample_1229
df_train.loc[df_train['A9']=='1900/1/9 7:00', 'A9'] = '7:00' # sample_937
df_train.loc[df_train['A9']=='700', 'A9'] = '7:00' # sample_496
df_train.loc[df_train['A11']=='1900/1/1 2:30', 'A11'] = '2:30' # sample_1067
df_train.loc[df_train['A11']==':30:00', 'A11'] = '00:30:00' # sample_130
df_train.loc[df_train['A16']=='1900/1/12 0:00', 'A16'] = '0:00' # sample_933
df_train.loc[df_train['A25']=='1900/3/10 0:00', 'A25'] = df_train['A25'].value_counts().index[0] # sample_1590
df_train['A25'] = df_train['A25'].astype(int)
df_train.loc[df_train['A26']=='1900/3/13 0:00', 'A26'] = '0:00' # sample_1350
# df_train.to_csv('fs.csv', index=False)
# Drop Outliers
df_train = df_train[df_train['A17']>89] # sample_1469
df_train = df_train[df_train['A22']>3.5] # sample_602
df_train = df_train[df_train['B1']!=3.5] # sample_102
df_train = df_train[df_train['B14']>290] # 13(2)
df_train = df_train[df_train['收率']>0.835] # 3
df_A = df_A[df_A['A19']<700] # sample_54
df_A = df_A[df_A['B14']>280] # 1, sample_1043
df_A = df_A[df_A['B14']<785] # 1, sample_316
df_B = df_B[df_B['收率']>0.823] # 1
df_C = df_C[df_C['B14']>320] # 3(1)
print('Processed:\nTrain samples: {}, Test samples: {},{},{}'.format(len(df_train), len(df_A), len(df_B), len(df_C)))
# Time Feature Processing
cols_time = ['A5', 'A7', 'A9', 'A11', 'A14', 'A16', 'A24', 'A26', 'B5', 'B7']
cols_period = ['A20', 'A28', 'B4', 'B9', 'B10', 'B11']
for col in cols_time:
df_train[col] = df_train[col].apply(get_time)
df_A[col] = df_A[col].apply(get_time)
df_B[col] = df_B[col].apply(get_time)
df_C[col] = df_C[col].apply(get_time)
df_submit[col] = df_submit[col].apply(get_time)
for col in cols_period:
df_train[col] = df_train[col].apply(get_period)
df_A[col] = df_A[col].apply(get_period)
df_B[col] = df_B[col].apply(get_period)
df_C[col] = df_C[col].apply(get_period)
df_submit[col] = df_submit[col].apply(get_period)
return df_train, df_A, df_B, df_C, df_submit
# additional features - Rev.2
def additional_features(df_train, df_A, df_B, df_C, df_submit):
train_len = len(df_train)
test_A_len = len(df_A)
test_B_len = len(df_B)
test_C_len = len(df_C)
df = df_train.append([df_A, df_B, df_C, df_submit])
# df.to_csv('df_all.csv', index=False)
# Ingredient Ratio
df['B12'] = df['B12'].fillna(df['B12'].value_counts().index[0])
df['B1'] = df['B1'].fillna(df['B1'].value_counts().index[0])
df['A2'] = df['A2'].fillna(0)
df['A3'] = df['A3'].fillna(0)
df['B14_B12'] = df['B14'] / df['B12']
df['B14_B1'] = df['B14'] / df['B1']
df['B14_A19'] = df['B14'] / df['A19']
df['B14_A4'] = df['B14'] / df['A4']
df['B14_A1'] = df['B14'] / df['A1']
df['b14_rate0'] = df['B14'] / (df['A1']+df['A2']+df['A3']+df['A4']+df['A19']+df['B1']+df['B12'])
df['b12_rate0'] = df['B12']/ (df['A1']+df['A2']+df['A3']+df['A4']+df['A19']+df['B1']+df['B12'])
df['b1_rate0'] = df['B1']/ (df['A1']+df['A2']+df['A3']+df['A4']+df['A19']+df['B1']+df['B12'])
df['a19_rate0'] = df['A19']/ (df['A1']+df['A2']+df['A3']+df['A4']+df['A19']+df['B1']+df['B12'])
df['a4_rate0'] = df['A4']/ (df['A1']+df['A2']+df['A3']+df['A4']+df['A19']+df['B1']+df['B12'])
# Time Related Diff
df['A21'] = df['A21'].fillna(df['A21'].value_counts().index[0])
df['A25'] = df['A25'].fillna(df['A25'].value_counts().index[0])
df['A27'] = df['A27'].fillna(df['A27'].value_counts().index[0])
df['B5'] = df['B5'].fillna(df['B5'].value_counts().index[0])
df['B8'] = df['B8'].fillna(df['B8'].value_counts().index[0])
df['A9-A5'] = df['A9'] - df['A5']
df['A10-A6'] = df['A10'] - df['A6']
df['A10-A6_A9-A5'] = df['A10-A6'] / df['A9-A5']
df['A12-A10'] = df['A12'] - df['A10']
df['A15-A12'] = df['A15'] - df['A12']
df['A17-A15'] = df['A17'] - df['A15']
df['A25-A21'] = df['A25'] - df['A21']
df['A27-A25'] = df['A27'] - df['A25']
df['B7-B5'] = df['B7'] - df['B5']
df['B8-B6'] = df['B8'] - df['B6']
df['B8-B6_B7-B5'] = df['B8-B6'] / df['B7-B5']
df_train = df[:train_len]
df_A = df[train_len : train_len+test_A_len]
df_B = df[train_len+test_A_len : train_len+test_A_len+test_B_len]
df_C = df[train_len+test_A_len+test_B_len : train_len+test_A_len+test_B_len+test_C_len]
df_submit = df[train_len+test_A_len+test_B_len+test_C_len:]
return df_train, df_A, df_B, df_C, df_submit
# Split Train and Valid
def dataset_split(df_train, df_A, df_B, df_C):
train_df = [df_train.append([df_B, df_C]), df_train.append([df_A, df_C]), df_train.append([df_A, df_B])]
test_df = [df_A, df_B, df_C]
return train_df, test_df
# LightGBM
def kfold_lightgbm(train_df, test_df, params, feats, df_submit, debug=False, B14_SI=True):
ids = ['A', 'B', 'C']
oof_preds = np.zeros((0,1))
oof_target = np.zeros((0,1))
sub_preds = np.zeros(df_submit.shape[0])
# k-fold, k=3
for id, train_xy, valid_xy in zip(ids, train_df, test_df):
train_x, train_y = train_xy[feats], train_xy['收率']
valid_x, valid_y = valid_xy[feats], valid_xy['收率']
test_x = df_submit[feats]
# Rev.4
if B14_SI:
df_b14_si = train_xy.groupby(['B14'], as_index=False)['收率'].agg({'gb_B14_tar'+'_mean': 'mean', 'gb_B14_tar'+'_max': 'max', 'gb_B14_tar'+'_min': 'min'})
train_x = pd.merge(train_x, df_b14_si, how='left', on=['B14'])
valid_x = pd.merge(valid_x, df_b14_si, how='left', on=['B14'])
test_x = pd.merge(test_x, df_b14_si, how='left', on=['B14'])
lgb_train = lgb.Dataset(train_x, label=train_y, free_raw_data=False)
lgb_test = lgb.Dataset(valid_x, label=valid_y, free_raw_data=False)
clf = lgb.train(params, lgb_train, valid_sets=[lgb_train, lgb_test], valid_names=['train', 'test'],
num_boost_round=10000, early_stopping_rounds=100, verbose_eval=False)
sub_preds += clf.predict(test_x, num_iteration=clf.best_iteration) / 3
oof_pred = clf.predict(valid_x, num_iteration=clf.best_iteration)
oof_preds = np.vstack((oof_preds, oof_pred.reshape(-1, 1)))
oof_target = np.vstack((oof_target, np.asarray(valid_y).reshape(-1, 1)))
score_test = mean_squared_error(oof_target, oof_preds)
print(f'Final CV: {score_test:.8f} !')
if debug:
output = pd.DataFrame()
output['target'] = list(oof_target)
output['pred'] = list(oof_preds)
output.to_csv('pred_lgb.csv')
return oof_preds, oof_target, sub_preds
# XGBoost
def kfold_xgboost(train_df, test_df, params, feats, df_submit, debug=False):
ids = ['A', 'B', 'C']
oof_preds = np.zeros((0,1))
oof_target = np.zeros((0,1))
sub_preds = np.zeros(df_submit.shape[0])
# k-fold, k=3
for id, train_xy, valid_xy in zip(ids, train_df, test_df):
train_x, train_y = train_xy[feats], train_xy['收率']
valid_x, valid_y = valid_xy[feats], valid_xy['收率']
test_x = df_submit[feats]
# Rev.4
df_b14_si = train_xy.groupby(['B14'], as_index=False)['收率'].agg({'gb_B14_tar'+'_mean': 'mean', 'gb_B14_tar'+'_max': 'max', 'gb_B14_tar'+'_min': 'min'})
train_x = pd.merge(train_x, df_b14_si, how='left', on=['B14'])
valid_x = pd.merge(valid_x, df_b14_si, how='left', on=['B14'])
test_x = pd.merge(test_x, df_b14_si, how='left', on=['B14'])
xgb_train = xgb.DMatrix(train_x, label=train_y)
xgb_test = xgb.DMatrix(valid_x, label=valid_y)
watchlist = [(xgb_train, 'train'), (xgb_test, 'valid_data')]
clf = xgb.train(params=params, dtrain=xgb_train, num_boost_round=10000, evals=watchlist, early_stopping_rounds=100, verbose_eval=False)
sub_preds += clf.predict(xgb.DMatrix(test_x), ntree_limit=clf.best_ntree_limit) / 3
oof_pred = clf.predict(xgb.DMatrix(valid_x), ntree_limit=clf.best_ntree_limit)
oof_preds = np.vstack((oof_preds, oof_pred.reshape(-1, 1)))
oof_target = np.vstack((oof_target, np.asarray(valid_y).reshape(-1, 1)))
score_test = mean_squared_error(oof_target, oof_preds)
print(f'Final CV: {score_test:.8f} !')
if debug:
output = pd.DataFrame()
output['target'] = list(oof_target)
output['pred'] = list(oof_preds)
output.to_csv('pred_xgb.csv')
return oof_preds, oof_target, sub_preds
def main(debug=False):
with timer('Train & Test'):
train, test_A, test_B, test_C, df_submit = train_test()
with timer('Additional Features'):
train_ad, test_A_ad, test_B_ad, test_C_ad, df_submit_ad = additional_features(train, test_A, test_B, test_C, df_submit)
with timer('Split Train & Test'):
feats_exclude = ['样本id', 'A1', 'A2', 'A3', 'A7', 'A8', 'A11', 'A13', 'A14', 'A16', 'A18', 'A23', 'A26', 'B2', 'B3', 'B13', '收率']
feats = [col for col in train.columns if col not in feats_exclude]
feats_ad = [col for col in train_ad.columns if col not in feats_exclude]
train_df, test_df = dataset_split(train, test_A, test_B, test_C)
train_df_ad, test_df_ad = dataset_split(train_ad, test_A_ad, test_B_ad, test_C_ad)
with timer('Run LightGBM_r0 with kfold'):
params = {'num_leaves': 47,
'min_data_in_leaf': 2,
'max_depth': -1,
'learning_rate': 0.05,
'boosting': 'gbdt',
'objective':'regression',
'feature_fraction': 1.0,
'bagging_freq': 1,
'bagging_fraction': 0.9,
'metric': 'mse',
'lambda_l1': 0,
'lambda_l2': 0,
'verbosity': -1,
'num_threads': 2}
oof_preds_lgb_r0, oof_target, sub_preds_lgb_r0 = kfold_lightgbm(train_df, test_df, params, feats, df_submit, debug=debug, B14_SI=False)
with timer('Run LightGBM_r4 with kfold'):
params = {'num_leaves': 47,
'min_data_in_leaf': 2,
'max_depth': -1,
'learning_rate': 0.1,
'boosting': 'gbdt',
'objective':'regression',
'feature_fraction': 0.8,
'bagging_freq': 1,
'bagging_fraction': 1.0,
'metric': 'mse',
'lambda_l1': 0,
'lambda_l2': 0.1,
'verbosity': -1,
'num_threads': 2}
oof_preds_lgb_r4, oof_target, sub_preds_lgb_r4 = kfold_lightgbm(train_df, test_df, params, feats, df_submit, debug=debug)
with timer('Run XGBoost_r4_IR+TD with kfold'):
params = {'learning_rate': 0.05,
'max_depth': 7,
'subsample': 0.8,
'colsample_bytree': 0.8,
'gamma': 0.0,
'min_child_weight': 1,
'reg_alpha': 0,
'reg_lambda': 2,
'objective': 'reg:linear',
'eval_metric': 'rmse',
'silent': True,
'seed': 0,
'nthread': 4}
oof_preds_xgb_r4IT, oof_target, sub_preds_xgb_r4IT = kfold_xgboost(train_df_ad, test_df_ad, params, feats_ad, df_submit_ad, debug=debug)
with timer('Stacking Above Models'):
oof_preds_avg = (oof_preds_lgb_r0+oof_preds_lgb_r4+oof_preds_xgb_r4IT)/3
score_test = mean_squared_error(oof_target, oof_preds_avg)
print(f'Final CV: {score_test:.8f} !')
sub_preds_avg = (sub_preds_lgb_r0+sub_preds_lgb_r4+sub_preds_xgb_r4IT)/3
df_submit['收率'] = sub_preds_avg
print('Special Value(B14) Processing...')
v = 385
print('Special Value(B14):', v)
submit_select = df_submit[df_submit['B14'] == v]['样本id'].index
df_submit.loc[submit_select, '收率'] = 0.878
print('Special Value(B14) Processed!')
df_submit[['样本id', '收率']][:-1].to_csv('submit_FuSai.csv', index=False, header=False)
df_submit[['样本id', '收率']][-1:].to_csv('submit_optimize.csv', index=False, header=False)
if __name__ == '__main__':
with timer('Full model run'):
main(debug=False)