6、可视化交叉验证

Posted 2023-05-15

参考技术A 6、可视化交叉验证

from sklearn.model_selection import (TimeSeriesSplit, KFold, ShuffleSplit,

                                    StratifiedKFold, GroupShuffleSplit,

                                    GroupKFold, StratifiedShuffleSplit)

import numpy as np

import matplotlib.pyplot as plt

from matplotlib.patches import Patch

np.random.seed(1338)

cmap_data = plt.cm.Paired

cmap_cv = plt.cm.coolwarm

n_splits = 4

plt.rcParams['font.sans-serif'] = ['SimHei']

plt.rcParams['axes.unicode_minus'] = False

np.random.seed(1338)

cmap_data = plt.cm.Paired

cmap_cv = plt.cm.coolwarm

n_splits = 4

# 生成类别/组数据

n_points = 100

X = np.random.randn(100, 10)

percentiles_classes = [.1, .3, .6]

y = np.hstack([[ii] * int(100 * perc)

              for ii, perc in enumerate(percentiles_classes)])

# 间隔均匀的组重复一次

groups = np.hstack([[ii] * 10 for ii in range(10)])

def visualize_groups(classes, groups, name):

    # 可视化数据集组

    fig, ax = plt.subplots()

    ax.scatter(range(len(groups)),  [.5] * len(groups), c=groups, marker='_',

              lw=50, cmap=cmap_data)

    ax.scatter(range(len(groups)),  [3.5] * len(groups), c=classes, marker='_',

              lw=50, cmap=cmap_data)

    ax.set(ylim=[-1, 5], yticks=[.5, 3.5],

          yticklabels=['Data\ngroup', 'Data\nclass'], xlabel="Sample index")

visualize_groups(y, groups, 'no groups')

plt.title("可视化数据", fontsize=15)

def plot_cv_indices(cv, X, y, group, ax, n_splits, lw=10):

    """为交叉验证对象的索引创建样本图."""

    # 为每个交叉验证分组生成训练/测试可视化图像

    for ii, (tr, tt) in enumerate(cv.split(X=X, y=y, groups=group)):

        # 与训练/测试组一起填写索引

        indices = np.array([np.nan] * len(X))

        indices[tt] = 1

        indices[tr] = 0

        # 可视化结果

        ax.scatter(range(len(indices)), [ii + .5] * len(indices),

                  c=indices, marker='_', lw=lw, cmap=cmap_cv,

                  vmin=-.2, vmax=1.2)

    # 将数据的分组情况和标签情况放入图像

    ax.scatter(range(len(X)), [ii + 1.5] * len(X),

              c=y, marker='_', lw=lw, cmap=cmap_data)

    ax.scatter(range(len(X)), [ii + 2.5] * len(X),

              c=group, marker='_', lw=lw, cmap=cmap_data)

    # 调整格式

    yticklabels = list(range(n_splits)) + ['class', 'group']

    ax.set(yticks=np.arange(n_splits+2) + .5, yticklabels=yticklabels,

          xlabel='Sample index', ylabel="CV iteration",

          ylim=[n_splits+2.2, -.2], xlim=[0, 100])

    ax.set_title(''.format(type(cv).__name__), fontsize=15)

    return ax

fig, ax = plt.subplots()

cv = KFold(n_splits)

plot_cv_indices(cv, X, y, groups, ax, n_splits)

fig, ax = plt.subplots()

cv = StratifiedKFold(n_splits)

plot_cv_indices(cv, X, y, groups, ax, n_splits)

cvs = [KFold, GroupKFold, ShuffleSplit, StratifiedKFold,

      GroupShuffleSplit, StratifiedShuffleSplit,

      TimeSeriesSplit]

for cv in cvs:

    this_cv = cv(n_splits=n_splits)

    fig, ax = plt.subplots(figsize=(6, 3))

    plot_cv_indices(this_cv, X, y, groups, ax, n_splits)

    ax.legend([Patch(color=cmap_cv(.8)), Patch(color=cmap_cv(.02))],

              ['Testing set', 'Training set'], loc=(1.02, .8))

    # Make the legend fit

    plt.tight_layout()

    fig.subplots_adjust(right=.7)

plt.title("可视化交叉验证\n StratifiedKGroupFold", fontsize=15)   

plt.show()

寻找模型最优参数多模型交叉验证可视化指标计算多模型对比可视化（系数图误差图混淆矩阵校正曲线ROC曲线AUCAccuracy特异度灵敏度PPVNPV）

使用randomsearchcv寻找模型最优参数、多模型交叉验证、可视化、指标计算、多模型对比可视化（系数图、误差图、classification_report、混淆矩阵、校正曲线、ROC曲线、AUC、Accuracy、特异度、灵敏度、PPV、NPV）

目录