为了账号安全,请及时绑定邮箱和手机立即绑定

树,回归,集成学习——让我们一起愉快地调参

2019.05.14 16:52 383浏览

完整代码如下:
特征X结构:X1_class/X1_num

import os
import pandas as pd 
import numpy  as np 

def train_data_reads(path):  
	data_directory  = path + "/data"
	#获取数据路径
	data_name_list  = os.listdir(data_directory)
	file_name       = data_name_list[0]
	#数据的路径:data_path
	data_path       = data_directory + "/" + file_name
	name,extension  = file_name.split(".")
	if extension == "csv":
		try:
			data = pd.read_csv(data_path,encoding = "gbk")
		except:
			data = pd.read_csv(data_path,encoding = "utf-8")
	elif extension == "txt":
		try:
			data = pd.read_csv(data_path,encoding = "gbk",sep = "\t")
		except:
			data = pd.read_csv(data_path,encoding = "utf-8",sep = "\t")
	else:
		data = pd.read_excel(data_path)

	return data 

def train_data_reprocess(data):

	#剔除重复值
	data = data.drop_duplicates()
	data = data.reset_index(drop = True)

	return data 

def feature_label_split(data):
	#获取dataFrame的名
	name_list = data.columns.values.tolist()
	label_name = name_list[len(name_list) - 1]

	#将数据中label为空的数据删除
	data = data[np.isnan(data[label_name]) == False]

	#拆分特征与标签
	x = data.drop(["ID",label_name],axis = 1)
	y = data[label_name]

	#补全特征中的缺失值
	feature_name_list = x.columns.values.tolist()
	class_name_list   = [name for name in feature_name_list if name.find("class") > 0]
	num_name_list     = [name for name in feature_name_list if name.find("num")   > 0]
	class_filled_df   = x[class_name_list].fillna("missing")
	num_filled_df     = x[num_name_list].fillna(data.mean())
	new_x             = pd.concat([class_filled_df,num_filled_df],axis = 1)

	return new_x,y


#将分类特征转换成哑变量
def dummy_variable_transform(x):
	#获取feature的列名
    columns_name = x.columns.values.tolist()
    for feature_name in columns_name:
        feature_name_split = feature_name.split("_", 1)
        name = feature_name_split[0]
        feature_type = feature_name_split[1]
        if feature_type == 'class':
            dummy_class = pd.get_dummies(x[feature_name], prefix=name, drop_first=True)
            x = x.drop(feature_name, axis=1).join(dummy_class)
    return x 

#对数据集X进行归一化
#线性回归对最大值,最小值敏感,思考一下,标准化Or归一化哪个更好
def data_MinMax(x):
	from sklearn.preprocessing import MinMaxScaler
	scaler = MinMaxScaler(feature_range = (0,1))
	scaler.fit(x)
	data = scaler.transform(x)
	return data 

def data_MinMax2(x):
	from sklearn.preprocessing import MinMaxScaler
	scaler = MinMaxScaler(feature_range = (-1,1))
	scaler.fit(x)
	data = scaler.transform(x)
	return data 

#为数据增加特征:多项式回归用的
def poly_data(x1):
	from sklearn.preprocessing import PolynomialFeatures
	poly = PolynomialFeatures(degree = 2)
	poly.fit(x1)
	x2 = poly.transform(x1)
	return x2

#划分训练集和测试集
def train_test_div(x,y,percent):
	from sklearn.model_selection import train_test_split
	x_train,x_test,y_train,y_test = train_test_split(x,y,test_size = percent)
	return x_train,x_test,y_train,y_test
	#train_test_split:先打乱顺序,然后进行分割


#1.线性回归预测
def lin_predict(x_train,x_test,y_train,y_test):
	from sklearn import linear_model
	from sklearn.linear_model import LinearRegression
	from sklearn.metrics import mean_squared_error,r2_score
	linreg = LinearRegression()
	linreg.fit(x_train,y_train)

	y_pred = linreg.predict(x_test)
	y_pred = list(map(lambda x: x if x >= 0 else 0,y_pred))
	#y小于0时,赋值为0
	y_pred = list(map(lambda x: x if x <= 10 else 10,y_pred))
	#y大于10时,赋值为10
	MAE = np.sum(np.absolute(y_pred - y_test)) / len(y_test)
	return MAE
	
#2.决策树预测
#决策树不需要变量变为哑变量
def tree_predict(x_train,x_test,y_train,y_test):
	from sklearn.tree import DecisionTreeRegressor
	reg = DecisionTreeRegressor(max_depth = 100,min_samples_split = 50,min_samples_leaf = 50)
	reg.fit(x_train,y_train)

	y_pred = reg.predict(x_test)
	y_pred = list(map(lambda x: x if x >= 0 else 0,y_pred))
	#y小于0时,赋值为0
	y_pred = list(map(lambda x: x if x <= 10 else 10,y_pred))
	#y大于10时,赋值为10
	MAE = np.sum(np.absolute(y_pred - y_test)) / len(y_test)
	return MAE


def rf_predict(x_train,x_test,y_train,y_test):
	from sklearn.ensemble import RandomForestRegressor
	rf  = RandomForestRegressor()
	rf.fit(x_train,y_train)

	y_pred = rf.predict(x_test)
	y_pred = list(map(lambda x: x if x >= 0 else 0,y_pred))
	#y小于0时,赋值为0
	y_pred = list(map(lambda x: x if x <= 10 else 10,y_pred))
	#y大于10时,赋值为10
	MAE = np.sum(np.absolute(y_pred - y_test)) / len(y_test)
	return MAE


def xgb_predict(x_train,x_test,y_train,y_test):
	import xgboost as xgb
	model_xgb = xgb.XGBRegressor(base_score=0.5, booster='gbtree', colsample_bylevel=0.7, colsample_bytree=0.7, gamma=0,
                                 learning_rate=0.05, max_delta_step=0, max_depth=6, min_child_weight=50, missing=None,
                                 n_estimators=350, n_jobs=-1, nthread=None, objective='reg:linear', random_state=2019,
                                 reg_alpha=0, reg_lambda=1, scale_pos_weight=1, seed=None, silent=True, subsample=1)
	model_xgb.fit(x_train, y_train)
	y_pred = model_xgb.predict(x_test)
	y_pred = list(map(lambda x: x if x >= 0 else 0,y_pred))
	y_pred = list(map(lambda x: x if x <= 10 else 10,y_pred))
	MAE = np.sum(np.absolute(y_pred - y_test)) / len(y_test)
	return MAE


def gbr_predict(x_train,x_test,y_train,y_test):
	from sklearn.ensemble import GradientBoostingRegressor
	model_gbr = GradientBoostingRegressor(alpha=0.6, criterion='friedman_mse', init=None, learning_rate=0.05, loss='ls',
                                          max_depth=3, max_features=None, max_leaf_nodes=None,
                                          min_impurity_decrease=0.0,
                                          min_impurity_split=None, min_samples_leaf=10, min_samples_split=2,
                                          min_weight_fraction_leaf=0.01, n_estimators=750, presort='auto',
                                          random_state=2019, subsample=0.7, verbose=0, warm_start=False)
	model_gbr.fit(x_train, y_train)
	y_pred = model_gbr.predict(x_test)
	y_pred = list(map(lambda x: x if x >= 0 else 0,y_pred))
	y_pred = list(map(lambda x: x if x <= 10 else 10,y_pred))
	MAE = np.sum(np.absolute(y_pred - y_test)) / len(y_test)
	return MAE


def xgb_predict2(x_train,x_test,y_train,y_test):
	import xgboost as xgb
	model_xgb = xgb.XGBRegressor(base_score=0.5, booster='gbtree', colsample_bylevel=0.7, colsample_bytree=0.7, gamma=0,
                                 learning_rate=0.05, max_delta_step=0, max_depth=6, min_child_weight=50, missing=None,
                                 n_estimators=600, n_jobs=-1, nthread=None, objective='reg:linear', random_state=2019,
                                 reg_alpha=0, reg_lambda=1, scale_pos_weight=1, seed=None, silent=True, subsample=0.9)
	model_xgb.fit(x_train, y_train)
	y_pred = model_xgb.predict(x_test)
	y_pred = list(map(lambda x: x if x >= 0 else 0,y_pred))
	y_pred = list(map(lambda x: x if x <= 10 else 10,y_pred))
	return y_pred


def gbr_predict2(x_train,x_test,y_train,y_test):
	from sklearn.ensemble import GradientBoostingRegressor
	model_gbr = GradientBoostingRegressor(alpha=0.6, criterion='friedman_mse', init=None, learning_rate=0.05, loss='ls',
                                          max_depth=3, max_features=None, max_leaf_nodes=None,
                                          min_impurity_decrease=0.0,
                                          min_impurity_split=None, min_samples_leaf=10, min_samples_split=2,
                                          min_weight_fraction_leaf=0.01, n_estimators=750, presort='auto',
                                          random_state=2019, subsample=0.9, verbose=0, warm_start=False)
	model_gbr.fit(x_train, y_train)
	y_pred = model_gbr.predict(x_test)
	y_pred = list(map(lambda x: x if x >= 0 else 0,y_pred))
	y_pred = list(map(lambda x: x if x <= 10 else 10,y_pred))
	return y_pred 


def main():
	path = "E:/AnaLinReg/Data"
	data = train_data_reads(path)
	data = train_data_reprocess(data)
	x,y = feature_label_split(data)
	x = x.iloc[0:30000,:]
	y = y.iloc[0:30000]
	x = dummy_variable_transform(x)
	x = x.astype(np.float64)
	x = data_MinMax(x)
	x_train,x_test,y_train,y_test = train_test_div(x,y,0.3)
	y1 = xgb_predict2(x_train,x_test,y_train,y_test)
	y2 = gbr_predict2(x_train,x_test,y_train,y_test)
	y =  [ ((y1[i] + y2[i]) / 2.0 ) for i in range(len(y1))]
	MAE = np.sum(np.absolute(y - y_test)) / len(y_test)
	print (MAE)
    
    

if __name__ == "__main__":
	main()

酸爽的调参过程
图片描述

点击查看更多内容
0人点赞

若觉得本文不错,就分享一下吧!

评论

相关文章推荐

正在加载中
意见反馈 邀请有奖 帮助中心 APP下载
官方微信

举报

0/150
提交
取消