-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodeltrainorg.py
176 lines (172 loc) · 7.41 KB
/
modeltrainorg.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
import pandas as pd
import numpy as np
import os
import lightgbm
from sklearn.model_selection import KFold
from sklearn.metrics import mean_squared_error
import matplotlib.pyplot as plt
import pdb
import scipy.io as io
plt.close("all")
#
#----------树模型直接预测每一天的流量，特征:year,month,day,weekday,WKD_TYP_CD-----
train_df=pd.read_csv('./train_v2.csv')
test_df=pd.read_csv('./test_v2_periods.csv')#按0.5h计算
test_day=pd.read_csv('./test_v2_day.csv')#按天计算
wkd_df=pd.read_csv('./wkd_v1.csv')
wkd_df=wkd_df.rename(columns={'ORIG_DT':'date'})
train_df=train_df.merge(wkd_df)
#处理特征
def get_frt(df):
df['WKD_TYP_CD']=df['WKD_TYP_CD'].map({'WN':0,'SN': 1, 'NH': 1, 'SS': 1, 'WS': 0})
df['date']=pd.to_datetime(df['date'])
df['dayofweek']=df['date'].dt.dayofweek+1
df['day']=df['date'].dt.day
df['month']=df['date'].dt.month
df['year']=df['date'].dt.year
df.drop(['date','post_id'],axis=1,inplace=True)
return df
#train
tmp=train_df[['date','post_id','amount']].groupby(['date','post_id'],sort=False)['amount'].sum()
train_day_df=tmp.reset_index()
train_day_df_A=train_day_df[train_day_df['post_id']=='A'].reset_index(drop=True)
train_day_df_B=train_day_df[train_day_df['post_id']=='B'].reset_index(drop=True)
train_day_df_A=train_day_df_A.merge(wkd_df)
train_day_df_B=train_day_df_B.merge(wkd_df)
#pdb.set_trace()
# 清洗数据
column_A_WN = train_day_df_A[(train_day_df_A['WKD_TYP_CD']=='WN')&(train_day_df_A['amount']==0)]['date']
column_A_WS = train_day_df_A[(train_day_df_A['WKD_TYP_CD']=='WS')&(train_day_df_A['amount']==0)]['date']
column_col = pd.concat([column_A_WN, column_A_WS], axis=0)
column_col = np.array(column_col).tolist()
# pdb.set_trace()
io.savemat('drop.mat',mdict={'my_drop':column_col})
train_day_df_A[(train_day_df_A['WKD_TYP_CD']=='WN')&(train_day_df_A['amount']==0)] = np.nan
train_day_df_A[(train_day_df_A['WKD_TYP_CD']=='WS')&(train_day_df_A['amount']==0)] = np.nan
train_day_df_A.dropna(how='any', axis=0, inplace=True)
# pdb.set_trace()
train_day_df_B[(train_day_df_B['WKD_TYP_CD']=='WN')&(train_day_df_B['amount']==0)] = np.nan
train_day_df_B[(train_day_df_B['WKD_TYP_CD']=='WS')&(train_day_df_B['amount']==0)] = np.nan
train_day_df_B.dropna(how='any', axis=0, inplace=True)
# pdb.set_trace()
train_day_df_A=get_frt(train_day_df_A)
train_day_df_B=get_frt(train_day_df_B)
train_day_df_A['amount']=train_day_df_A['amount']/1e4
train_day_df_B['amount']=train_day_df_B['amount']/1e4
#test
tmp=test_day[['date','post_id']]
test_day_df=tmp
# pdb.set_trace()
test_day_df_A=test_day_df[test_day_df['post_id']=='A'].reset_index(drop=True)
test_day_df_B=test_day_df[test_day_df['post_id']=='B'].reset_index(drop=True)
test_day_df_A=test_day_df_A.merge(wkd_df)
test_day_df_B=test_day_df_B.merge(wkd_df)
test_day_df_A=get_frt(test_day_df_A)
test_day_df_B=get_frt(test_day_df_B)
#训练集和测试集
print(train_day_df_A.shape,test_day_df_A.shape)#(1035, 6) (30, 5)
def lgb_cv(train_x, train_y, test_x):
predictors = list(train_x.columns)
train_x = train_x.values
test_x = test_x.values
folds = 10
seed = 2021
kf = KFold(n_splits=folds, shuffle=True, random_state=seed)
train = np.zeros((train_x.shape[0]))
test = np.zeros((test_x.shape[0]))
test_pre = np.zeros((folds, test_x.shape[0]))
test_pre_all = np.zeros((folds, test_x.shape[0]))
cv_scores = []
tpr_scores = []
cv_rounds = []
for i, (train_index, test_index) in enumerate(kf.split(train_x, train_y)):
tr_x = train_x[train_index]
tr_y = train_y[train_index]
te_x = train_x[test_index]
te_y = train_y[test_index]
train_matrix = lightgbm.Dataset(tr_x, label=tr_y)
test_matrix = lightgbm.Dataset(te_x, label=te_y)
params = {
'boosting_type': 'gbdt',
'objective': 'regression',
'metrics':'mean_squared_error',
'num_leaves': 2 ** 5-1,
'feature_fraction': 0.8,
'bagging_fraction': 0.8,
'learning_rate': 0.05,
'seed': 2021,
'nthread': 8,
'verbose': -1,
}
num_round = 4000
early_stopping_rounds = 100
if test_matrix:
model = lightgbm.train(params, train_matrix, num_round, valid_sets=test_matrix, verbose_eval=200,
#feval=tpr_eval_score,
early_stopping_rounds=early_stopping_rounds
)
print("\n".join(("%s: %.2f" % x) for x in list(sorted(zip(predictors, model.feature_importance("gain")),
key=lambda x: x[1],reverse=True))[:10]))
importance_list=[ x[0] for x in list(sorted(zip(predictors, model.feature_importance("gain")),
key=lambda x: x[1],reverse=True))]
#print(importance_list)
pre = model.predict(te_x, num_iteration=model.best_iteration)#
pred = model.predict(test_x, num_iteration=model.best_iteration)#
train[test_index] = pre
test_pre[i, :] = pred
cv_scores.append(mean_squared_error (te_y, pre))
cv_rounds.append(model.best_iteration)
test_pre_all[i, :] = pred
#
print("cv_score is:", cv_scores)
use_mean=True
if use_mean:
test[:] = test_pre.mean(axis=0)
else:
pass
#
print("val_mean:" , np.mean(cv_scores))
print("val_std:", np.std(cv_scores))
return train, test, test_pre_all, np.mean(cv_scores),importance_list
if __name__=="__main__":
select_frts=['WKD_TYP_CD','year','month','day','dayofweek']
train_df=train_day_df_A#训练集A
train_df=train_df[(train_df['year']==2020) & (train_df['month']>5)].reset_index(drop=True)
test_df=test_day_df_A#测试集A
train_x = train_df[select_frts].copy()
train_y = train_df['amount']
test_x = test_df[select_frts].copy()
print(train_x.shape,train_y.shape,test_x.shape)
lgb_train, lgb_test, sb, cv_scores, importance_list = lgb_cv(train_x, train_y, test_x)
lgb_test_A=[item if item>0 else 0 for item in lgb_test]
#
train_df=train_day_df_B#训练集B
train_df=train_df[(train_df['year']==2020) & (train_df['month']>5)].reset_index(drop=True)
test_df=test_day_df_B#测试集B
train_x = train_df[select_frts].copy()
train_y = train_df['amount']
test_x = test_df[select_frts].copy()
print(train_x.shape,train_y.shape,test_x.shape)
lgb_train, lgb_test, sb, cv_scores, importance_list = lgb_cv(train_x, train_y, test_x)
lgb_test_B=[item if item>0 else 0 for item in lgb_test]
print(np.mean(lgb_test_A),np.sum(lgb_test_A),np.mean(lgb_test_B),np.sum(lgb_test_B))
#
pre_A=np.array(lgb_test_A)
pre_B=np.array(lgb_test_B)
test_day=pd.read_csv('./test_v2_day.csv')#按天计算
pre_day=[]
for i in range(31):
pre_day.append(pre_A[i]*1e4)
if(i+2)%7==6 or (i+2)%7==0:
pre_day.append(0)
else:
pre_day.append(pre_B[i]*1e4)
test_day['amount']=pre_day
#
if not os.path.exists('submitTreenew/'):
os.makedirs('submitTreenew/')
f=open('submitTreenew/test_day_day.txt','w')
f.write('Date'+','+'Post_id'+','+'Predict_amount'+'\n')
for _,date,post_id,amount in test_day.itertuples():
f.write(date+','+post_id+','+str(int(amount))+'\n')
f.close()