【集成學習】lightgbm使用案例

github地址

#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Sat Mar 31 21:19:09 2018

@author: hello4720
"""
import numpy as np
import pandas as pd
import lightgbm as lgb
from sklearn import metrics
from sklearn.model_selection import train_test_split

### 讀取數據
print("載入數據")
dataset1 = pd.read_csv('G:/ML/ML_match/IJCAI/data3.22/3.22ICJAI/data/7_train_data1.csv')
dataset2 = pd.read_csv('G:/ML/ML_match/IJCAI/data3.22/3.22ICJAI/data/7_train_data2.csv')
dataset3 = pd.read_csv('G:/ML/ML_match/IJCAI/data3.22/3.22ICJAI/data/7_train_data3.csv')
dataset4 = pd.read_csv('G:/ML/ML_match/IJCAI/data3.22/3.22ICJAI/data/7_train_data4.csv')
dataset5 = pd.read_csv('G:/ML/ML_match/IJCAI/data3.22/3.22ICJAI/data/7_train_data5.csv')

dataset1.drop_duplicates(inplace=True)
dataset2.drop_duplicates(inplace=True)
dataset3.drop_duplicates(inplace=True)
dataset4.drop_duplicates(inplace=True)
dataset5.drop_duplicates(inplace=True)

### 數據合並
print("數據合並")
trains = pd.concat([dataset1,dataset2],axis=0)
trains = pd.concat([trains,dataset3],axis=0)
trains = pd.concat([trains,dataset4],axis=0)

online_test = dataset5

### 數據拆分
print("數據拆分")
train_xy,offline_test = train_test_split(trains, test_size = 0.2,random_state=21)
train,val = train_test_split(train_xy, test_size = 0.2,random_state=21)

print("訓練集")
y = train.is_trade                                                  # 訓練集標簽
X = train.drop(['instance_id','is_trade'],axis=1)                   # 訓練集特征矩陣

print("驗證集")
val_y = val.is_trade                                                # 驗證集標簽
val_X = val.drop(['instance_id','is_trade'],axis=1)                 # 驗證集特征矩陣

print("測試集")
offline_test_X=offline_test.drop(['instance_id','is_trade'],axis=1) # 線下測試特征矩陣
online_test_X=online_test.drop(['instance_id'],axis=1)              # 線上測試特征矩陣

### 數據轉換
lgb_train = lgb.Dataset(X, y, free_raw_data=False)
lgb_eval = lgb.Dataset(val_X, val_y, reference=lgb_train,free_raw_data=False)

### 開始訓練
print('設置參數')
params = {
            'boosting_type': 'gbdt',
            'boosting': 'dart',
            'objective': 'binary',
            'metric': 'binary_logloss',

            'learning_rate': 0.01,
            'num_leaves':25,
            'max_depth':3,

            'max_bin':10,
            'min_data_in_leaf':8,

            'feature_fraction': 0.6,
            'bagging_fraction': 1,
            'bagging_freq':0,

            'lambda_l1': 0,
            'lambda_l2': 0,
            'min_split_gain': 0
}

print("開始訓練")
gbm = lgb.train(params,                     # 參數字典
                lgb_train,                  # 訓練集
                num_boost_round=2000,       # 迭代次數
                valid_sets=lgb_eval,        # 驗證集
                early_stopping_rounds=30)   # 早停系數
### 線下預測
print ("線下預測")
preds_offline = gbm.predict(offline_test_X, num_iteration=gbm.best_iteration) # 輸出概率
offline=offline_test[['instance_id','is_trade']]
offline['preds']=preds_offline
offline.is_trade = offline['is_trade'].astype(np.float64)
print('log_loss', metrics.log_loss(offline.is_trade, offline.preds))

### 線上預測
print("線上預測")
preds_online =  gbm.predict(online_test_X, num_iteration=gbm.best_iteration)  # 輸出概率
online=online_test[['instance_id']]
online['preds']=preds_online
online.rename(columns={'preds':'predicted_score'},inplace=True)
online.to_csv("./data/20180405.txt",index=None,sep=' ')

### 保存模型
from sklearn.externals import joblib
joblib.dump(gbm,'gbm.pkl')

### 特征選擇
df = pd.DataFrame(X.columns.tolist(), columns=['feature'])
df['importance']=list(gbm.feature_importance())
df = df.sort_values(by='importance',ascending=False)
df.to_csv("./data/feature_score_20180405.csv",index=None,encoding='gbk')