-
Notifications
You must be signed in to change notification settings - Fork 105
/
Copy pathtrain_cv.py
94 lines (75 loc) · 3.86 KB
/
train_cv.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
import numpy as np
import pandas as pd
import scipy as sp
import lightgbm as lgb
import gc
import datetime
import random
import scipy.special as special
from sklearn.cross_validation import train_test_split
from sklearn.cross_validation import StratifiedKFold
rawpath='C:\\final\\'
temppath='C:\\final\\temp\\'
iapath='C:\\final\\temp\\installedactions\\'
def logloss(act, preds):
epsilon = 1e-15
preds = sp.maximum(epsilon, preds)
preds = sp.minimum(1 - epsilon, preds)
ll = sum(act * sp.log(preds) + sp.subtract(1, act) * sp.log(sp.subtract(1, preds)))
ll = ll * -1.0 / len(act)
return ll
def getTrainVal(X_train, scope=(28, 29), val_type='30', seed=1000):
if val_type == '30':
X_val = X_train.loc[X_train['day'] == 30, :]
X_train = X_train.loc[(X_train['day'] >= scope[0]) & (X_train['day'] <= scope[1]), :]
elif val_type == '73':
X_train = X_train.loc[(X_train['day'] >= scope[0]) & (X_train['day'] <= scope[1]), :]
X_train, X_val, y_train, y_val = train_test_split(X_train, X_train['label'], test_size=0.3, random_state=seed)
return X_train, X_val
t_start = datetime.datetime.now()
X_loc_train=pd.read_csv(temppath+'2_smooth.csv')
print('load train over...')
X_loc_test=pd.read_csv(temppath+'2_test_smooth.csv')
print('load test over...')
##########################################################CV预测时30号验证效果不好，而其中一折做验证来提前停止，后面删除30天数据
X_loc_train, X_loc_val = getTrainVal(X_loc_train, scope=(28, 29), val_type='30', seed=1000)
drop = ['label', 'day']
y_loc_train = X_loc_train.loc[:, 'label']
X_loc_train.drop(drop, axis=1, inplace=True)
# y_loc_val = X_loc_val.loc[:, 'label']
# X_loc_val.drop(drop, axis=1, inplace=True)
res = X_loc_test.loc[:, ['instanceID']]
X_loc_test.drop(['instanceID'], axis=1, inplace=True)
X_loc_test.drop(drop, axis=1, inplace=True)
gc.collect()
print('preprocess over...', X_loc_train.shape)
##########################################################比赛只用了lightGBM单模型
X_loc_train=X_loc_train.values
y_loc_train=y_loc_train.values
# X_loc_val=X_loc_val.values
# y_loc_val=y_loc_val.values
X_loc_test=X_loc_test.values
##########################################################交叉预测，实际上是stacking第一层做的操作
# 利用不同折数加参数，特征，样本（随机数种子）扰动，再加权平均得到最终成绩
model = lgb.LGBMClassifier(boosting_type='gbdt', num_leaves=29, max_depth=-1, learning_rate=0.1, n_estimators=10000,
max_bin=425, subsample_for_bin=50000, objective='binary', min_split_gain=0,
min_child_weight=5, min_child_samples=10, subsample=1, subsample_freq=1,
colsample_bytree=1, reg_alpha=3, reg_lambda=5, seed=1000, nthread=-1, silent=True)
del X_loc_val
skf=list(StratifiedKFold(y_loc_train, n_folds=10, shuffle=True, random_state=1024))
for i, (train, test) in enumerate(skf):
print("Fold", i)
model.fit(X_loc_train[train], y_loc_train[train], eval_metric='logloss',eval_set=[(X_loc_train[train], y_loc_train[train]), (X_loc_train[test], y_loc_train[test])],early_stopping_rounds=100)
preds= model.predict_proba(X_loc_test, num_iteration=-1)[:, 1]
print('mean:', preds.mean())
res['prob_%s' % str(i)] = preds
#model._Booster.save_model('lgb.model')
#model = xgb.Booster(model_file='xgb.model')
#平均或者加权的方式有很多种，台大三傻的比赛分享里有一个利用sigmoid反函数来平均的方法效果不错
now = datetime.datetime.now()
now = now.strftime('%m-%d-%H-%M')
print(now)
res.sort_values("instanceID", ascending=True, inplace=True)
res.to_csv(rawpath+"%s.csv" % now, index=False)
t_end = datetime.datetime.now()
print('training time: %s' % ((t_end - t_start).seconds/60))