lightGBM使用案例
#!/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
### 读取数据
print("载入数据")
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)
### 数据合并
print("数据合并")
trains = pd.concat([dataset1,dataset2],axis=0)
trains = pd.concat([trains,dataset3],axis=0)
trains = pd.concat([trains,dataset4],axis=0)
online_test = dataset5
### 数据拆分
print("数据拆分")
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("训练集")
y = train.is_trade # 训练集标签
X = train.drop(['instance_id','is_trade'],axis=1) # 训练集特征矩阵
print("验证集")
val_y = val.is_trade # 验证集标签
val_X = val.drop(['instance_id','is_trade'],axis=1) # 验证集特征矩阵
print("测试集")
offline_test_X=offline_test.drop(['instance_id','is_trade'],axis=1) # 线下测试特征矩阵
online_test_X=online_test.drop(['instance_id'],axis=1) # 线上测试特征矩阵
### 数据转换
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)
### 开始训练
print('设置参数')
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("开始训练")
gbm = lgb.train(params, # 参数字典
lgb_train, # 训练集
num_boost_round=2000, # 迭代次数
valid_sets=lgb_eval, # 验证集
early_stopping_rounds=30) # 早停系数
### 线下预测
print ("线下预测")
preds_offline = gbm.predict(offline_test_X, num_iteration=gbm.best_iteration) # 输出概率
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))
### 线上预测
print("线上预测")
preds_online = gbm.predict(online_test_X, num_iteration=gbm.best_iteration) # 输出概率
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')
### 特征选择
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')调参案例
lightgbm使用leaf_wise tree生长策略，leaf_wise_tree的优点是收敛速度快，缺点是容易过拟合。
# lightgbm关键参数
# -*- coding: utf-8 -*-
"""
# 作者：wanglei5205
# 邮箱：[email protected]
# 博客：http://cnblogs.com/wanglei5205
# github：http://github.com/wanglei5205
"""
### 导入模块
import numpy as np
import pandas as pd
import lightgbm as lgb
from sklearn import metrics
### 载入数据
print('载入数据')
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')
print('数据去重')
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)
print('数据合并')
trains = pd.concat([dataset1,dataset2],axis=0)
trains = pd.concat([trains,dataset3],axis=0)
trains = pd.concat([trains,dataset4],axis=0)
online_test = dataset5
### 数据拆分(训练集+验证集+测试集)
print('数据拆分')
from sklearn.model_selection import train_test_split
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)
# 训练集
y_train = train.is_trade # 训练集标签
X_train = train.drop(['instance_id','is_trade'],axis=1) # 训练集特征矩阵
# 验证集
y_val = val.is_trade # 验证集标签
X_val = val.drop(['instance_id','is_trade'],axis=1) # 验证集特征矩阵
# 测试集
offline_test_X = offline_test.drop(['instance_id','is_trade'],axis=1) # 线下测试特征矩阵
online_test_X = online_test.drop(['instance_id'],axis=1) # 线上测试特征矩阵
### 数据转换
print('数据转换')
lgb_train = lgb.Dataset(X_train, y_train, free_raw_data=False)
lgb_eval = lgb.Dataset(X_val, y_val, reference=lgb_train,free_raw_data=False)
### 设置初始参数--不含交叉验证参数
print('设置参数')
params = {
'boosting_type': 'gbdt',
'objective': 'binary',
'metric': 'binary_logloss',
}
### 交叉验证(调参)
print('交叉验证')
min_merror = float('Inf')
best_params = {}
# 准确率
print("调参1：提高准确率")
for num_leaves in range(20,200,5):
for max_depth in range(3,8,1):
params['num_leaves'] = num_leaves
params['max_depth'] = max_depth
cv_results = lgb.cv(
params,
lgb_train,
seed=2018,
nfold=3,
metrics=['binary_error'],
early_stopping_rounds=10,
verbose_eval=True
)
mean_merror = pd.Series(cv_results['binary_error-mean']).min()
boost_rounds = pd.Series(cv_results['binary_error-mean']).argmin()
if mean_merror < min_merror:
min_merror = mean_merror
best_params['num_leaves'] = num_leaves
best_params['max_depth'] = max_depth
params['num_leaves'] = best_params['num_leaves']
params['max_depth'] = best_params['max_depth']
# 过拟合
print("调参2：降低过拟合")
for max_bin in range(1,255,5):
for min_data_in_leaf in range(10,200,5):
params['max_bin'] = max_bin
params['min_data_in_leaf'] = min_data_in_leaf
cv_results = lgb.cv(
params,
lgb_train,
seed=42,
nfold=3,
metrics=['binary_error'],
early_stopping_rounds=3,
verbose_eval=True
)
mean_merror = pd.Series(cv_results['binary_error-mean']).min()
boost_rounds = pd.Series(cv_results['binary_error-mean']).argmin()
if mean_merror < min_merror:
min_merror = mean_merror
best_params['max_bin']= max_bin
best_params['min_data_in_leaf'] = min_data_in_leaf
params['min_data_in_leaf'] = best_params['min_data_in_leaf']
params['max_bin'] = best_params['max_bin']
print("调参3：降低过拟合")
for feature_fraction in [0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]:
for bagging_fraction in [0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]:
for bagging_freq in range(0,50,5):
params['feature_fraction'] = feature_fraction
params['bagging_fraction'] = bagging_fraction
params['bagging_freq'] = bagging_freq
cv_results = lgb.cv(
params,
lgb_train,
seed=42,
nfold=3,
metrics=['binary_error'],
early_stopping_rounds=3,
verbose_eval=True
)
mean_merror = pd.Series(cv_results['binary_error-mean']).min()
boost_rounds = pd.Series(cv_results['binary_error-mean']).argmin()
if mean_merror < min_merror:
min_merror = mean_merror
best_params['feature_fraction'] = feature_fraction
best_params['bagging_fraction'] = bagging_fraction
best_params['bagging_freq'] = bagging_freq
params['feature_fraction'] = best_params['feature_fraction']
params['bagging_fraction'] = best_params['bagging_fraction']
params['bagging_freq'] = best_params['bagging_freq']
print("调参4：降低过拟合")
for lambda_l1 in [0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]:
for lambda_l2 in [0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]:
for min_split_gain in [0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]:
params['lambda_l1'] = lambda_l1
params['lambda_l2'] = lambda_l2
params['min_split_gain'] = min_split_gain
cv_results = lgb.cv(
params,
lgb_train,
seed=42,
nfold=3,
metrics=['binary_error'],
early_stopping_rounds=3,
verbose_eval=True
)
mean_merror = pd.Series(cv_results['binary_error-mean']).min()
boost_rounds = pd.Series(cv_results['binary_error-mean']).argmin()
if mean_merror < min_merror:
min_merror = mean_merror
best_params['lambda_l1'] = lambda_l1
best_params['lambda_l2'] = lambda_l2
best_params['min_split_gain'] = min_split_gain
params['lambda_l1'] = best_params['lambda_l1']
params['lambda_l2'] = best_params['lambda_l2']
params['min_split_gain'] = best_params['min_split_gain']
print(best_params)
### 训练
params['learning_rate']=0.01
lgb.train(
params, # 参数字典
lgb_train, # 训练集
valid_sets=lgb_eval, # 验证集
num_boost_round=2000, # 迭代次数
early_stopping_rounds=50 # 早停次数
)
### 线下预测
print ("线下预测")
preds_offline = lgb.predict(offline_test_X, num_iteration=lgb.best_iteration) # 输出概率
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))
### 线上预测
print("线上预测")
preds_online = lgb.predict(online_test_X, num_iteration=lgb.best_iteration) # 输出概率
online=online_test[['instance_id']]
online['preds']=preds_online
online.rename(columns={'preds':'predicted_score'},inplace=True) # 更改列名
online.to_csv("./data/20180405.txt",index=None,sep=' ') # 保存结果
### 保存模型
from sklearn.externals import joblib
joblib.dump(lgb,'lgb.pkl')
### 特征选择
df = pd.DataFrame(X_train.columns.tolist(), columns=['feature'])
df['importance']=list(lgb.feature_importance()) # 特征分数
df = df.sort_values(by='importance',ascending=False) # 特征排序
df.to_csv("./data/feature_score_20180331.csv",index=None,encoding='gbk') # 保存分数