机器学习实验作业

根据本章所学内容，选取任一机器学习算法（包括神经网络），编写对应代码。并简述梯度下降算法的原理和流程、什么是过拟合及其解决方案、如何根据混淆矩阵计算准确率等指标。

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn.metrics import precision_recall_curve, confusion_matrix, recall_score, precision_score, accuracy_score, \
    f1_score, roc_curve, auc
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import GridSearchCV
import seaborn as sns

from 逻辑回归 import List_score, AUC

pd.set_option('display.max_columns', None)


def data_read():
    df = pd.read_csv(r"C:\Users\zhang\Desktop\附件133.csv", encoding="gbk")
    df.rename(columns={
   'Y(1=default, 0=non-default)': 'Y'}, inplace=True)
    return df


def get_category_bin(df, col, binsnum, labels, qcut=False):
    ## 这块没考虑null和""，之后优化代码可以考虑
    if qcut:
        localdf = pd.qcut(df[col], q=binsnum, labels=labels)
    else:
        localdf = pd.cut(df[col], bins=binsnum, labels=labels)  # equal-length cut  可以根据指定分界点对连续数据进行分箱处理
    localdf = pd.DataFrame(localdf)
    name = col + '_' + 'bin'
    localdf[name] = localdf[col]
    df = df.join(localdf[name])  ## 新增加一列 连接进去
    df = df.drop(i, axis=1)
    df[name] = df[name].astype(object)
    return df


def category_continue_separation(df, feature_names):
    categorical_var = []
    numerical_var = []
    if 'Y' in feature_names:
        feature_names.remove('Y')
    if '编号' in feature_names:
        feature_names.remove('编号')
    ##先判断类型，如果是int或float就直接作为连续变量
    numerical_var = list(df[feature_names].select_dtypes(
        include=['int', 'float', 'int32', 'float32', 'int64', 'float64']).columns.values)
    categorical_var = numerical_var
    return categorical_var, numerical_var


def data_analysis_plot(categorical_var, df, start, mid, row, col, showlabel=False):
    p = 1
    plt.figure(figsize=(20, 18))
    for i in categorical_var[start:mid]:
        df_var = "df_" + i
        df_var = pd.crosstab(df[i], df["Y"])
        df_var["bad_rate"] = round(df_var[1] / (df_var[1] + df_var[0]), 4)
        # print('df_var', df_var)
        # print(df_var.index)
        # print(df_var.columns)
        # print(p)

        # 添加子区间绘制，
        ax1 = plt.subplot(row, col, p)  # 子区空间为n行n列

        year = df_var.index  # 横坐标
        data1 = df_var[0]  # 纵坐标 好用户
        data2 = df_var[1]  # 纵坐标 坏用户
        data3 = df_var["bad_rate"]

        # 先得到year长度, 再得到下标组成列表
        x = range(len(year))
        bar_width = 0.3

        # 好坏用户柱状图
        ax1.bar(x, data1, width=bar_width, color='#3A669C', label="好用户")
        # 向右移动0.2, 柱状条宽度为0.2
        ax1.bar([i + bar_width for i in x], data2, width=bar_width, color='#C0504D', label="违约用户")

        # 底部汉字移动到两个柱状条中间(本来汉字是在左边蓝色柱状条下面, 向右移动0.1)
        plt.xticks([i + bar_width / 2 for i in x], year)
        if p == 1 or p == col + 1:
            ax1.set_ylabel('用户数量', size=10)
        ax1.set_xlabel(i, size=10)

        text_heiht = 3
        # 为每个条形图添加数值标签
        for x1, y1 in enumerate(data1):
            ax1.text(x1, y1 + text_heiht, y1, ha='center', fontsize=8)
        for x2, y2 in enumerate(data2):
            ax1.text(x2 + bar_width, y2 + text_heiht, y2, ha='center', fontsize=8)

        # bad_rate线 共用纵坐标轴
        ax2 = ax1.twinx()
        p3 = ax2.plot([i + bar_width / 2 for i in x], data3, color="gray", linestyle='--', label="坏用户占比")

        if p == 3 or p == 6:
            ax2.set_ylabel("坏用户占比"