vlambda博客
学习文章列表

# calibration_curve()函数格式from sklearn.calibration import calibration_curve
fraction_of_positives, mean_predicted_value = calibration_curve(y_train, prob_true,n_bins=20)# 注：prob_true需为概率。 # calibration_curve函数计算校准曲线，体现预测概率与真实分类的差距    # 第一个参数是样本真实分类，第二个参数是样本预测概率，n_bins参数表示把概率从小到大分成几个区间    # 返回的第一个数组是每一个区间中样本为真的比例，第二个数组是每一个区间中样本的平均预测概率    # 当曲线在虚线下方时，表示预测为真>实际为真，当曲线在虚线上方时，表示预测为真<实际为真

# Dataset# 导入库import pandas as pd# 导入数据data = pd.read_csv(r"C:\Users\lenovo\Desktop\04_16_disease.csv")# 数据描述print(data.describe())# 检查数据中是否有缺失值import numpy as npprint(np.isnan(data).any())data[data.isnull().values==True]# 取每列的缺失值个数：n_missings_each_col = data.apply(lambda x: x.isnull().sum())print(n_missings_each_col)

# 确定特征和目标X = data.iloc[:, 34:].valuesy = data.iloc[:, 0].values# 划分测试集和训练集from sklearn.model_selection import train_test_splitX_train, X_test, y_train, y_test = train_test_split(X, y,test_size=0.3, random_state=42)

# 导入逻辑回归模型from sklearn.linear_model import LogisticRegression
lr = LogisticRegression(solver='lbfgs')
# 绘制校准曲线from sklearn.calibration import calibration_curve
n_bins = 10prob_true = lr.predict_proba(X_train)[:, 1] # results are tuples of (Pr(False), Pr(True))fraction_of_positives, mean_predicted_value = calibration_curve(y_train, prob_true, n_bins=n_bins)
fig, ax = plt.subplots(figsize=(12, 6))ax.plot(mean_predicted_value, fraction_of_positives, marker='o')ax.set_xlabel('Average Predicted Value')ax.set_ylabel('Proportion of Positive Records')ax.plot([0, 1], [0, 1], linestyle='--')

# 绘制预测概率分组的柱状图fig, ax = plt.subplots(figsize=(12, 6))ax.hist(prob_true, bins=10)

# Dataset# 导入库import pandas as pd# 导入数据data = pd.read_csv(r"C:\Users\lenovo\Desktop\04_16_disease.csv")# 数据描述print(data.describe())# 检查数据中是否有缺失值import numpy as npprint(np.isnan(data).any())data[data.isnull().values==True]# 取每列的缺失值个数：n_missings_each_col = data.apply(lambda x: x.isnull().sum())print(n_missings_each_col)

# 确定特征和目标X = data.iloc[:, 34:].valuesy = data.iloc[:, 0].values# 划分测试集和训练集from sklearn.model_selection import train_test_splitX_train, X_test, y_train, y_test = train_test_split(X, y,test_size=0.3, random_state=42)

# Calibration curves (二分类问题)import numpy as np
from sklearn.svm import LinearSVC

class NaivelyCalibratedLinearSVC(LinearSVC):    """LinearSVC with `predict_proba` method that naively scales    `decision_function` output."""
    def fit(self, X, y):        super().fit(X, y)        df = self.decision_function(X)        self.df_min_ = df.min()        self.df_max_ = df.max()
    def predict_proba(self, X):        """Min-max scale output of `decision_function` to [0,1].""" # 将分数限制在0~1之间，与上面的概率取值范围保持一致，才能同步比较        df = self.decision_function(X) # decision_function函数返回每一个样本属于每一个类别的置信分数，这个分数是根据样本到分隔超平面的距离计算的        calibrated_df = (df - self.df_min_) / (self.df_max_ - self.df_min_)        proba_pos_class = np.clip(calibrated_df, 0, 1)        proba_neg_class = 1 - proba_pos_class        proba = np.c_[proba_neg_class, proba_pos_class]        return proba

from sklearn.calibration import CalibrationDisplayfrom sklearn.ensemble import RandomForestClassifierfrom sklearn.linear_model import LogisticRegressionfrom sklearn.naive_bayes import GaussianNB
# Create classifiers# 实例化四种分类算法：逻辑回归、朴素贝叶斯、线性支持向量分类器、随机森林lr = LogisticRegression(solver='lbfgs')gnb = GaussianNB()svc = NaivelyCalibratedLinearSVC(C=1.0)rfc = RandomForestClassifier()
clf_list = [    (lr, "Logistic"),    (gnb, "Naive Bayes"),    (svc, "SVC"),    (rfc, "Random forest"),]

import matplotlib.pyplot as pltfrom matplotlib.gridspec import GridSpec
fig = plt.figure(figsize=(15, 15))gs = GridSpec(4, 2) # 用于划分子图colors = plt.cm.get_cmap("Dark2")
ax_calibration_curve = fig.add_subplot(gs[:2, :2])calibration_displays = {}for i, (clf, name) in enumerate(clf_list): clf.fit(X_train, y_train) display = CalibrationDisplay.from_estimator( clf, X_test, y_test, n_bins=10, name=name, ax=ax_calibration_curve, color=colors(i), ) calibration_displays[name] = display
ax_calibration_curve.grid()ax_calibration_curve.set_ylabel("fraction of positives")ax_calibration_curve.set_ylim([-0.05, 1.05])ax_calibration_curve.legend(loc="lower right")ax_calibration_curve.set_title('Calibration plots (Reliability curve)') # calibration_curve函数计算校准曲线，体现预测概率与真实分类的差距 # 第一个参数是样本真实分类，第二个参数是样本预测概率，n_bins参数表示把概率从小到大分成几个区间 # 返回的第一个数组是每一个区间中样本为真的比例，第二个数组是每一个区间中样本的平均预测概率 # 当曲线在虚线下方时，表示预测为真>实际为真，当曲线在虚线上方时，表示预测为真<实际为真


# Add histogramgrid_positions = [(2, 0), (2, 1), (3, 0), (3, 1)]for i, (_, name) in enumerate(clf_list): row, col = grid_positions[i] ax = fig.add_subplot(gs[row, col])
 ax.hist( calibration_displays[name].y_prob, range=(0, 1), bins=10,# 在子图上绘制将0~1平均分为10个区间，每一个区间的样本数目 label=name, color=colors(i), ) ax.set(title=name, xlabel="Mean predicted probability", ylabel="Count")
plt.tight_layout()plt.show()# 设置子图的标签、轴刻度范围、legend样例集、使用紧凑的排列，避免子图超出绘图区域不显示

plt.tight_layout()

# 其他评价指标from collections import defaultdict
import pandas as pd
from sklearn.metrics import (    precision_score,    recall_score,    f1_score,    brier_score_loss,    log_loss,    roc_auc_score,)
scores = defaultdict(list)for i, (clf, name) in enumerate(clf_list):    clf.fit(X_train, y_train)    y_prob = clf.predict_proba(X_test)    y_pred = clf.predict(X_test)    scores["Classifier"].append(name)
    for metric in [brier_score_loss, log_loss]:        score_name = metric.__name__.replace("_", " ").replace("score", "").capitalize()        scores[score_name].append(metric(y_test, y_prob[:, 1]))
    for metric in [precision_score, recall_score, f1_score, roc_auc_score]:        score_name = metric.__name__.replace("_", " ").replace("score", "").capitalize()        scores[score_name].append(metric(y_test, y_pred))
    score_df = pd.DataFrame(scores).set_index("Classifier")    score_df.round(decimals=3)
score_df