import os
import gc
import time
import math
import psutil
import itertools
import numpy as np
import pandas as pd
from tqdm import tqdm
import lightgbm as lgb
import warnings
warnings.filterwarnings('ignore')
from sklearn.preprocessing import LabelEncoder
from sklearn.metrics import f1_score,roc_auc_score
from sklearn.model_selection import StratifiedKFold
def summary(f):
def wrap(*args, **kwargs):
time1 = time.time()
p = psutil.Process(os.getpid())
m0 = p.memory_info()[0] / 2. ** 30
ret = f(*args, **kwargs)
m1 = p.memory_info()[0] / 2. ** 30
delta = m1 - m0
sign = '+' if delta >= 0 else '-'
delta = math.fabs(delta)
time2 = time.time()
print('')
print('process :'.ljust(20) , f'{f.__name__}')
print('run time :'.ljust(20) , f'{np.round(time2-time1, 2)} s')
print('memory usage :'.ljust(20) , f'{m1:.1f}GB({sign}{delta:.1f}GB)')
return ret
return wrap
@summary
def load_data():
data_path = os.getcwd().replace('code','A榜给选手数据\\')
result_path = os.getcwd().replace('code','result\\')
train = pd.read_csv(data_path+'train_set.csv').drop(['X38','X27'],axis=1)
label = pd.read_csv(data_path+'train_label.csv')
test = pd.read_csv(data_path+'result_predict_A.csv').drop(['X38','X27'],axis=1)
df = pd.concat([train,test],axis=0).reset_index(drop=True)
df = df.merge(label,on='user_id',how='left')
return df
@summary
def fill_na(df):
cat_na_cols = ['X3','X5','X28','X29','X30','X31']
num_na_cols = ['X6','X7','X8','X9','X10','X11','X12','X13','X14','X15','X16','X17','X18','X19','X20','X21','X22','X23','X32','X33','X34','X35','X36']
tmp = df[['user_id']]
for col in num_na_cols:
tmp = pd.concat([tmp,df[col].fillna(df[col].median())],axis=1)
for col in cat_na_cols:
tmp = pd.concat([tmp,df[col].fillna(df[col].mode())],axis=1)
df.drop(cat_na_cols+num_na_cols,axis=1,inplace=True)
df = df.merge(tmp,on='user_id',how='left')
return df
@summary
def feature_engineering(df):
df['basic_combine'] = df['X1'].map(str)+df['X2'].map(str)+df['X3'].map(str)+df['X4'].map(str)+df['X5'].map(str)
df['kuandai_combine'] = df['X24'].map(str)+df['X25'].map(str)+df['X26'].map(str)
df['qianyue_combine'] = df['X28'].map(str)+df['X29'].map(str)+df['X30'].map(str)+df['X31'].map(str)
df['else_combine'] = df['X37'].map(str)+df['X39'].map(str)+df['X40'].map(str)+df['X41'].map(str)+df['X42'].map(str)+df['X43'].map(str)
features = [['X6','X7','X8'],['X9','X10','X11'],['X12','X13','X14'],['X18','X19','X20'],['X21','X22','X23']]
for fea in features:
df[f'{fea[0]}_{fea[1]}_{fea[2]}_std'] = df[fea].std(1)
df[f'{fea[0]}_{fea[1]}_{fea[2]}_max'] = df[fea].max(1)
df[f'{fea[0]}_{fea[1]}_{fea[2]}_min'] = df[fea].min(1)
df[f'{fea[0]}_{fea[1]}_sub'] = df[fea[0]] - df[fea[1]]
df[f'{fea[0]}_{fea[2]}_sub'] = df[fea[0]] - df[fea[2]]
df.loc[df[fea[0]] <= df[fea[1]],f'{fea[0]}_{fea[1]}_mark'] = 0
df.loc[df[fea[0]] > df[fea[1]],f'{fea[0]}_{fea[1]}_mark'] = 1
df.loc[df[fea[0]] <= df[fea[2]],f'{fea[0]}_{fea[2]}_mark'] = 0
df.loc[df[fea[0]] > df[fea[2]],f'{fea[0]}_{fea[2]}_mark'] = 1
features = ['X18','X19','X20','X21','X22','X23']
for fea in features:
df.loc[df[fea] == 0,f'{fea}_mark'] = 1
df.loc[df[fea] > 0,f'{fea}_mark'] = 0
mark_cols = [col for col in df.columns if 'mark' in col]
df['total_mark'] = 0
for col in mark_cols:
df['total_mark'] += df[col]
df.drop(mark_cols,axis=1,inplace=True)
gc.collect()
le = LabelEncoder()
for col in [col for col in df.columns if df[col].dtype == 'object']:
df[col] = le.fit_transform(df[col].astype(str))
return df
@summary
def model_f1(ta,te):
res = [0 for _ in range(len(test))]
fea = [c for c in ta.columns if c not in ['user_id','product_no','label']]
kf = StratifiedKFold(n_splits=5,shuffle=True,random_state=2)
f1 = []
threshold = 0.25
for ta_idx , val_idx in kf.split(ta,ta['label']):
X_ta, X_val = ta[fea].iloc[ta_idx], ta[fea].iloc[val_idx],
y_ta, y_val = np.array(ta['label'])[ta_idx], np.array(ta['label'])[val_idx]
model = lgb.LGBMClassifier(num_leaves=64,max_depth=13,n_estimators=10000,learning_rate=0.07,verbose=-1,metric='auc')
model.fit(X_ta, y_ta, eval_set = [(X_val,y_val)], early_stopping_rounds=200, verbose=500)
y_pred = model.predict_proba(X_val)[:,1]
res += model.predict_proba(te[fea])[:,1] / 5
y_pred[y_pred>threshold], y_pred[y_pred<=threshold] = 1, 0
f1.append(f1_score(y_val,y_pred))
imp = pd.Series(model.feature_importances_, fea).sort_values(ascending=False)
print('\nmean_f1:',np.around(np.mean(f1),3))
return res,imp
if __name__ == '__main__':
df = load_data()
df = fill_na(df)
df = feature_engineering(df)
df_ = df.copy()
ta,te = df_[~df_['label'].isna()], df_[df_['label'].isna()]
te['label'],imp = model_f1(ta,te)
res = te.copy()
res.loc[res['label']>0.24,'label'] = 1
res.loc[res['label']<0.24,'label'] = 0
res[['user_id','label']].to_csv(result_path+'sub_0.24.csv',index=False)
网友评论