-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy path2022xunfei_product_sale_predict.py
306 lines (259 loc) · 14.2 KB
/
2022xunfei_product_sale_predict.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
import os
import math
import random
import pandas as pd
import numpy as np
from matplotlib.pyplot import plot, show
import lightgbm as lgb
import xgboost as xgb
import catboost as cat
from sklearn.metrics import mean_squared_error
from sklearn.model_selection import KFold,StratifiedKFold
from sklearn.preprocessing import LabelEncoder
from sklearn.linear_model import Ridge
import warnings
warnings.simplefilter(action='ignore', category=FutureWarning)
warnings.filterwarnings('ignore')
def seed_everything(seed):
random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
np.random.seed(seed)
def calc_acc(y_true,y_pred):
acc=0
howmanay=len(y_true)//209
for i in range(howmanay):
start=i*209
end=(i+1)*209
target=y_true[start:end].tolist()
predict=y_pred[start:end].tolist()
for j in range(len(target)):
t=target[j]
p=predict[j]
if t == 0:
continue
acc+=(1-np.abs(p-t)/t)*(t/np.sum(target))
return acc/howmanay
def get_data():
seed_everything(2022)
fusai = True
root_path='../input/2022xunfei-product-sales-data/'
product_month_order_train_path = root_path+f'product_month_order_train{"_fusai" if fusai else ""}.csv'
product_need_train_path = root_path+f'product_need_train{"_fusai" if fusai else ""}.csv'
product_month_order_test_path = root_path+f'product_month_order_test{"_fusai" if fusai else ""}.csv'
product_need_test_path = root_path+f'product_need_test{"_fusai" if fusai else ""}.csv'
need_df=pd.read_csv(product_need_train_path)
order_df=pd.read_csv(product_month_order_train_path).rename(columns={'year':'year_id','month':'month_id'})
need_df['date']=pd.to_datetime(need_df['date'])
need_df=pd.DataFrame(
[{'product_id':product_id,'date':date} for product_id in need_df.product_id.unique()
for date in pd.date_range(need_df.date.min(),need_df.date.max())]).merge(need_df,on=['product_id','date'],how='left')
need_df=need_df.fillna(0)
need_df['year_id']=need_df['date'].dt.year
need_df['month_id']=need_df['date'].dt.month
need_df=need_df.groupby(['product_id','year_id','month_id'],as_index=False).agg({'is_sale_day':['nunique','max','sum'],'label':[('','sum')]})
need_df.columns=['_'.join(i) if i[1] != '' else i[0] for i in need_df.columns.to_list()]
train_df=need_df.merge(order_df,on=['product_id','year_id','month_id'],how='left')
need_df=pd.read_csv(product_need_test_path)
order_df=pd.read_csv(product_month_order_test_path).rename(columns={'year':'year_id','month':'month_id'})
need_df['date']=pd.to_datetime(need_df['date'])
need_df=pd.DataFrame(
[{'product_id':product_id,'date':date} for product_id in need_df.product_id.unique()
for date in pd.date_range(need_df.date.min(),need_df.date.max())]).merge(need_df,on=['product_id','date'],how='left')
need_df=need_df.fillna(0)
need_df['year_id']=need_df['date'].dt.year
need_df['month_id']=need_df['date'].dt.month
need_df=need_df.groupby(['product_id','year_id','month_id'],as_index=False).agg({'is_sale_day':['nunique','max','sum']})
need_df.columns=['_'.join(i) if i[1] != '' else i[0] for i in need_df.columns.to_list()]
test_df=need_df.merge(order_df,on=['product_id','year_id','month_id'],how='left')
data=pd.concat([train_df,test_df],ignore_index=True)
data=data.sort_values(by=['year_id','month_id', 'product_id']).reset_index(drop=True)
data['time_id'] = list(map(lambda x,y:(x-2018)*12+y-1,data['year_id'],data['month_id']))
data['jidu_id']=((data['month_id']-1)/3+1).map(int)
data['is_sale_day_max']=data['is_sale_day_max'].astype(int)
data['type']=LabelEncoder().fit_transform(data['type'])
data['label'] = np.log1p(data['label'])
# data['label'] = list(map(lambda x : x if x==np.NAN else math.log(x+1,2),data['label']))
feats_cols=data.columns.to_list()
for i in range(1, 17):
for f in ['label', 'order', 'start_stock', 'end_stock']:
data[f+'_shift_%d'%i] = data.groupby('product_id')[f].shift(i+2)
if i <= 12:
feats_cols.append(f+'_shift_%d'%i)
for i in [3, 6, 12]:
for f in ['label', 'order', 'start_stock', 'end_stock']:
data[f+'_mean_%d'%i] = data[[f+'_shift_%d'%i for i in range(1, i+1)]].mean(axis=1)
data[f+'_std_%d'%i] = data[[f+'_shift_%d'%i for i in range(1, i+1)]].std(axis=1)
data[f+'_median_%d'%i] = data[[f+'_shift_%d'%i for i in range(1, i+1)]].median(axis=1)
feats_cols.extend([f+'_mean_%d'%i,f+'_std_%d'%i,f+'_median_%d'%i])
category_cols=[
'product_id','year_id','month_id','time_id','jidu_id','is_sale_day_nunique','is_sale_day_max','type'
]
print('data.shape：',data.shape)
return data,feats_cols,category_cols
def train_model_with_nfold(data,train,test,feat_cols,feats_cols,category_cols,
lgb_params=None,xgb_params=None,cat_params=None,model_types=['lgb','xgb','cat'],fold_num=5,
seeds=[2022],stratified=True,num_boost_round=10000,early_stopping_rounds=200,verbose=200,
un_select_cols=[]
):
score_lgb = np.zeros(fold_num)
score_xgb = np.zeros(fold_num)
score_cat = np.zeros(fold_num)
score = np.zeros(fold_num)
oof_lgb = np.zeros(len(train))
oof_xgb = np.zeros(len(train))
oof_cat = np.zeros(len(train))
oof = np.zeros(len(train))
pred_y = pd.DataFrame()
for seed in seeds:
for model_type in model_types:
if stratified:
kf = StratifiedKFold(n_splits=fold_num, shuffle=True, random_state=seed)
else:
kf = KFold(n_splits=fold_num, shuffle=True, random_state=seed)
if model_type == 'cat':
feat_cols = [col for col in feats_cols if col not in un_select_cols+['label']]
np.random.shuffle(feat_cols)
if model_type == 'xgb':
if 1001 in train['product_id'].values:
LE=LabelEncoder()
for col in category_cols:
LE.fit(data[col].astype(str))
train[col]=LE.transform(train[col].astype(str))
test[col]=LE.transform(test[col].astype(str))
else:
train = data[data['label'].notna()].reset_index(drop=True)
test = data[data['label'].isna()].reset_index(drop=True)
for fold, (train_idx, val_idx) in enumerate(kf.split(train[feat_cols], train['product_id'])):
print(f'-----------------fold：{fold+1} -----------------seed：{seed} -----------------model_type：{model_type}')
if model_type == 'lgb':
lgb_params['seed']=seed
tra = lgb.Dataset(train.loc[train_idx, feat_cols],train.loc[train_idx, 'label'])
val = lgb.Dataset(train.loc[val_idx, feat_cols],train.loc[val_idx, 'label'])
model = lgb.train(lgb_params, tra, valid_sets=[val], num_boost_round=num_boost_round,categorical_feature=category_cols,
callbacks=[lgb.early_stopping(early_stopping_rounds), lgb.log_evaluation(verbose)])
score_lgb[fold]=model.best_score['valid_0']['rmse'] / len(seeds)
score[fold]=model.best_score['valid_0']['rmse'] / len(seeds) / len(model_types)
oof_lgb[val_idx] += model.predict(train.loc[val_idx, feat_cols], num_iteration=model.best_iteration) / len(seeds)
oof[val_idx] += model.predict(train.loc[val_idx, feat_cols], num_iteration=model.best_iteration) / len(seeds) / len(model_types)
pred_y[f'fold{fold}_seed{seed}_{model_type}'] = model.predict(test[feat_cols], num_iteration=model.best_iteration)
elif model_type == 'xgb':
xgb_params['seed']=seed
train_matrix = xgb.DMatrix(train.loc[train_idx, feat_cols] , label=train.loc[train_idx, 'label'])
valid_matrix = xgb.DMatrix(train.loc[val_idx, feat_cols] , label=train.loc[val_idx, 'label'])
watchlist = [(train_matrix, 'train'),(valid_matrix, 'eval')]
model = xgb.train(xgb_params, train_matrix, num_boost_round=num_boost_round, evals=watchlist, verbose_eval=verbose, early_stopping_rounds=early_stopping_rounds)
score_xgb[fold]=model.best_score / len(seeds)
score[fold]=model.best_score / len(seeds) / len(model_types)
oof_xgb[val_idx] += model.predict(valid_matrix,iteration_range=(0,model.best_iteration+1)) / len(seeds)
oof[val_idx] += model.predict(valid_matrix,iteration_range=(0,model.best_iteration+1)) / len(seeds) / len(model_types)
pred_y[f'fold{fold}_seed{seed}_{model_type}'] = model.predict(xgb.DMatrix(test[feat_cols]),iteration_range=(0,model.best_iteration+1))
else:
cat_params['random_seed']=seed
model = cat.CatBoostRegressor(num_boost_round=num_boost_round, **cat_params)
trn_x = train.loc[train_idx, feat_cols]
trn_y = train.loc[train_idx, 'label']
val_x = train.loc[val_idx, feat_cols]
val_y = train.loc[val_idx, 'label']
model.fit(trn_x, trn_y, eval_set=(val_x, val_y),cat_features=category_cols, use_best_model=True, verbose=verbose)
score_cat[fold] = model.best_score_['validation']['RMSE'] / len(seeds)
score[fold] = model.best_score_['validation']['RMSE'] / len(seeds) / len(model_types)
oof_cat[val_idx] += model.predict(val_x) / len(seeds)
oof[val_idx] += model.predict(val_x) / len(seeds) / len(model_types)
pred_y[f'fold{fold}_seed{seed}_{model_type}'] = model.predict(test[feat_cols])
print(f'score_lgb={score_lgb.mean()}\nscore_xgb={score_xgb.mean()}\nscore_cat={score_cat.mean()}\nscore={score.mean()}')
return oof_lgb,oof_xgb,oof_cat,oof,pred_y
def RidgeCV(train,oof_lgb,oof_xgb,oof_cat,pred_y,fold_num=5,seed=2022):
oof=pd.DataFrame([oof_lgb,oof_xgb,oof_cat]).T
oof.columns=['lgb','xgb','cat']
oof['product_id']=train['product_id']
oof['label']=train['label']
y=pd.DataFrame([pred_y[[i for i in pred_y.columns if 'lgb' in i]].mean(1),pred_y[[i for i in pred_y.columns if 'xgb' in i]].mean(1),pred_y[[i for i in pred_y.columns if 'cat' in i]].mean(1)]).T
y.columns=['lgb','xgb','cat']
kf = StratifiedKFold(n_splits=fold_num, shuffle=True, random_state=seed)
oof_res=np.zeros(len(oof))
prediction=np.zeros(len(y))
for fold, (train_idx, val_idx) in enumerate(kf.split(oof[['lgb','xgb','cat']], oof['product_id'])):
train_x=oof.loc[train_idx,['lgb','xgb','cat']]
train_y=oof.loc[train_idx,'label']
valid_x=oof.loc[val_idx,['lgb','xgb','cat']]
valid_y=oof.loc[val_idx,'label']
model = Ridge(random_state=2022)
model.fit(train_x,train_y)
oof_res[val_idx]=model.predict(valid_x)
prediction+=model.predict(y)/fold_num
return oof_res,prediction
def main():
data,feats_cols,category_cols=get_data()
train = data[data['label'].notna()].reset_index(drop=True)
test = data[data['label'].isna()].reset_index(drop=True)
un_select_cols=[]
feat_cols = [col for col in feats_cols if col not in un_select_cols+['label']]
np.random.shuffle(feat_cols)
print('feat_cols.len：',len(feat_cols))
#lgb
lgb_params={
'lambda_l2': 3.1841775244883856,
'learning_rate': 0.011388441559132073,
'max_depth': 28,
'min_child_weight': 3.404010799108265,
'num_leaves': 77,
'feature_pre_filter': False,
'boosting_type': 'gbdt',
'objective': 'regression',
'metric': 'rmse',
'seed': 2022,
'n_jobs': -1,
'verbose': -1
}
# xgb
xgb_params = {
'booster': 'gbtree',
'objective': 'reg:squarederror',
'eval_metric': 'rmse',
'eta': 0.07660224004250459,
'max_depth': 9,
'min_child_weight': 11.269262917397498
}
# cat
cat_params = {
'learning_rate': 0.3,
'loss_function':'RMSE',
'depth': 6,
'l2_leaf_reg': 3,
'od_type': 'Iter',
'od_wait': 200,
'allow_writing_files': False
}
oof_lgb,oof_xgb,oof_cat,oof,pred_y=train_model_with_nfold(data,train,test,feat_cols,feats_cols,category_cols,
lgb_params=lgb_params,xgb_params=xgb_params,cat_params=cat_params,model_types=['lgb','xgb','cat'],fold_num=5,
seeds=[2022],stratified=True,num_boost_round=10000,early_stopping_rounds=200,verbose=200,
un_select_cols=[]
)
oof_res,prediction=RidgeCV(train,oof_lgb,oof_xgb,oof_cat,pred_y,fold_num=5,seed=2022)
train['target_weight'] = train['label'] / train.groupby(['year_id', 'month_id'])['label'].transform('sum')
train['oof'] = np.expm1(oof_res)
train['oof'] = train['oof'].map(lambda x: x if x >= 0 else 0)
train['oof'] = train['oof'].round()
train['label'] = np.expm1(train['label'])
print(np.expm1(prediction).sum())
print(math.sqrt(mean_squared_error(train['label'],train['oof'])))
# train['oof'] = list(map(lambda x : x if x==np.NAN else (2**(x))-1, oof))
# train['label'] = list(map(lambda x : x if x==np.NAN else (2**(x))-1, train['label']))
score1 = np.sum((1 - np.abs(train['label']-train['oof']) / (train['label'])) * train['target_weight']) / (len(train)//209)
print(score1)
print(np.mean(np.abs(train['label']-train['oof'])/(train['label']+1)))
print(calc_acc(train['label'],train['oof']))
plot(train['label'])
plot(train['oof'])
show()
test['label'] = np.expm1(prediction)
# test['label'] = list(map(lambda x : x if x==np.NAN else (2**(x))-1, pred_y.mean(axis=1).values))
sub=data[data['label'].isna()].reset_index(drop=True)[['product_id','year_id','month_id','label']].copy()
sub['label'] = test['label'].map(lambda x: x if x >= 0 else 0)
sub['month']=sub['year_id'].astype(str)+'-'+sub['month_id'].apply(lambda x:f'{x:02}')
sub=sub[['month','product_id','label']]
sub['label']=sub['label'].round()
return sub
if __name__ == '__main__':
sub = main()
sub.to_csv('submit.csv',index=False)