GBDT+LR学习记录- 9.8模型在测试数据集表现（上）

GBDT+LR学习记录

9.8模型在测试数据集表现（上）

新建check.py文件
新建函数check_core（后来03：30改名为run_check）
check_core我们这里与lr相比只重新选取一下打分函数即可
新建函数get_test_data（同lr）
新建函数run_check_core（同lr）
下面新建定义一下打分函数predict_by_tree
新建get_auc、get_accuracy函数
写main函数
最上边引入模块from future import division

此时check.py如下

# -*- coding: utf-8 -*-

"""
==================================================
   File Name：     check

   Description :
==================================================
"""

from __future__ import division
import sys
import numpy as np
import xgboost as xgb

from scipy.sparse import csc_matrix  #稀疏矩阵，方便计算
import math

#9-8和lr测试模型在数据集上的表现一致
def get_test_data(test_file, feature_num_file):
    """读取测试文件"""
    total_feature_num = 103
    # total_feature_num = utils.get_feature_num(feature_num_file)
    test_label = np.genfromtxt(test_file, dtype=np.float32, delimiter=',', usecols=-1)
    total_feature_list = range(total_feature_num)
    test_feature = np.genfromtxt(test_file, dtype=np.float32, delimiter=',', usecols=total_feature_list)
    return test_feature, test_label

# 定义下打分函数
def predict_by_tree(test_feature, tree_model):  # 2个输入：测试特征+模型
    """predict by gbdt model"""
    predict_list = tree_model.predict(xgb.DMatrix(test_feature))  # 转化数据结构
    return predict_list

def get_auc(predict_list, test_label):
    """
    auc得分
    :param predict_list: model predict score list
    :param test_label:  label of test data
    auc = sum(pos_index) - pos_num(pos_num+1)/2  / (pos_num *neg_num)

    """
    total_list = []
    for index in range(len(predict_list)):
        predict_score = predict_list[index]
        label = test_label[index]
        total_list.append((label, predict_score))
    scorted_total_list = sorted(total_list, key=lambda ele: ele[1])
    neg_num = 0
    pos_num = 0
    count = 1
    total_pos_index = 0

    for value in scorted_total_list:
        label, predict_score = value
        if label == 0:
            neg_num += 1
        else:
            pos_num += 1
            total_pos_index += count
        count += 1
    auc_score = (total_pos_index - (pos_num) * (pos_num + 1) / 2) / (pos_num * neg_num + 1)

    print("auc: %5f " % (auc_score))


def get_accuracy(predict_list, test_label):
    """
    predict_list:model predict list score list
    test_label: lable of test data
    :return:
    """
    right_num = 0
    score_thr = 0.5
    for index in range(len(predict_list)):
        predict_score = predict_list[index]
        if predict_score >= score_thr:
            predict_label = 1
        else:
            predict_label = 0

        if predict_label == test_label[index]:
            right_num += 1
    total_num = len(predict_list)
    accuracy_score = right_num / total_num
    print("accuracy: %5f    " % (accuracy_score))


#同lr
def run_check_core(test_feature, test_label, model, score_func):
    """评分函数"""
    predict_list = score_func(test_feature, model)
    get_auc(predict_list, test_label)
    get_accuracy(predict_list, test_label)


#9-8
# 3个参数——测试文件，gbdt模型，记录所有特征维度的文件
def run_check(test_file, tree_model_file, feature_num_file):
    """gbdt模型预测"""
    test_feature, test_label = get_test_data(test_file, feature_num_file)  # 获得测试的feature和label
    # 加载模型
    test_model = xgb.Booster(model_file=tree_model_file)
    # 打分函数  #zin即我们这里与lr相比只重新选取一下打分函数即可
    run_check_core(test_feature, test_label, test_model, predict_by_tree)


if __name__ == "__main__":
    # 视频写的如下
    #     run_check("data/gbdt_test_file", "data/xgb.model", "data/gbdt_feature_num", )
    run_check("data/gbdt_test_file", "data/gbdt.model", "data/gbdt_feature_num", )

但是运行报错

D:\develop\Anaconda3\python.exe "D:/develop/PyCharm 2018.3.5_workspace/p_recommendation_mys