示例简介
本示例介绍如何绘制不同分类器的分类概率。我们在拥有3个类别的数据集上,使用如下分类模型进行多分类:
- 支持向量分类器(SVM Classification, 简称SVC)
- 线性SVC在默认情况下不是概率分类器,但在此示例中启用了内置校准选项(
probability=True
)
- 线性SVC在默认情况下不是概率分类器,但在此示例中启用了内置校准选项(
- 逻辑回归(LogisticRegression,简称LR),尝试了3种参数配置:
- 采用L1正则化
- 采用L2正则化和One-Vs-Rest模式,One-Vs-Rest简称OvR,即做多分类时,将1个类的样本作为正例,其他类的样本作为负例。
- 使用One-Vs-Rest的逻辑回归不是开箱即用的多类分类器。在下文中可用看到,与其他模型估计量相比,在将第2类和第3类分离时会有一些问题。
- 采用L2正则化和multinomial模式,multinomial即直接做多分类。
- 高斯过程分类(GaussianProcessClassifier,简称GPC)。
代码实现[Python]
# -*- coding: utf-8 -*-
print(__doc__)
# Author: Alexandre Gramfort
# License: BSD 3 clause
import matplotlib.pyplot as plt
import numpy as np
from sklearn.metrics import accuracy_score
from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC
from sklearn.gaussian_process import GaussianProcessClassifier
from sklearn.gaussian_process.kernels import RBF
from sklearn import datasets
iris = datasets.load_iris()
X = iris.data[:, 0:2] # we only take the first two features for visualization
y = iris.target
n_features = X.shape[1]
C = 10
kernel = 1.0 * RBF([1.0, 1.0]) # for GPC
# 创建5个不同的分类器.
classifiers = {
'L1 logistic': LogisticRegression(C=C, penalty='l1',
solver='saga',
multi_class='multinomial',
max_iter=10000),
'L2 logistic (Multinomial)': LogisticRegression(C=C, penalty='l2',
solver='saga',
multi_class='multinomial',
max_iter=10000),
'L2 logistic (OvR)': LogisticRegression(C=C, penalty='l2',
solver='saga',
multi_class='ovr',
max_iter=10000),
'Linear SVC': SVC(kernel='linear', C=C, probability=True,
random_state=0),
'GPC': GaussianProcessClassifier(kernel)
}
n_classifiers = len(classifiers)
plt.figure(figsize=(3 * 2, n_classifiers * 2))
plt.subplots_adjust(bottom=.2, top=.95)
xx = np.linspace(3, 9, 100)
yy = np.linspace(1, 5, 100).T
xx, yy = np.meshgrid(xx, yy)
Xfull = np.c_[xx.ravel(), yy.ravel()]
for index, (name, classifier) in enumerate(classifiers.items()):
classifier.fit(X, y)
y_pred = classifier.predict(X)
accuracy = accuracy_score(y, y_pred)
print("Accuracy (train) for %s: %0.1f%% " % (name, accuracy * 100))
# View probabilities:
probas = classifier.predict_proba(Xfull)
n_classes = np.unique(y_pred).size
for k in range(n_classes):
plt.subplot(n_classifiers, n_classes, index * n_classes + k + 1)
plt.title("Class %d" % k)
if k == 0:
plt.ylabel(name)
imshow_handle = plt.imshow(probas[:, k].reshape((100, 100)),
extent=(3, 9, 1, 5), origin='lower')
plt.xticks(())
plt.yticks(())
idx = (y_pred == k)
if idx.any():
plt.scatter(X[idx, 0], X[idx, 1], marker='o', c='w', edgecolor='k')
ax = plt.axes([0.15, 0.04, 0.7, 0.05])
plt.title("Probability")
plt.colorbar(imshow_handle, cax=ax, orientation='horizontal')
plt.show()
代码执行
代码运行时间大约:0分1.409秒。
运行代码输出的文本内容如下,如上文所述OvR的LR不是天然的多分类器,所以效果相对较差。
Accuracy (train) for L1 logistic: 83.3% Accuracy (train) for L2 logistic (Multinomial): 82.7% Accuracy (train) for L2 logistic (OvR): 79.3% Accuracy (train) for Linear SVC: 82.0% Accuracy (train) for GPC: 82.7%
运行代码输出的图片内容如下,图中白色圆点表示样本,背景色的深浅表示概率的强度(见最后一张图,颜色越浅(黄)概率越大,颜色越深(深蓝)概率越小)。我们再次可以看到,对于OvR的逻辑回归,效果不太好——Class 1和Class 2的样本对应的概率强度比较低!
源码下载
- Python版源码文件: plot_classification_probability.py
- Jupyter Notebook版源码文件: plot_classification_probability.ipynb