sklearn-第八节（可视化）

1 简介

本次主要通过使用Scikit-Plot的模块来介绍机器学习的相关可视化，Scikit-Plot主要包括以下几个部分：

• estimators：用于绘制各种算法
• metrics：用于绘制机器学习的onfusion matrix, ROC AUC curves, precision-recall curves等曲线
• cluster：主要用于绘制聚类
• decomposition：主要用于绘制PCA降维

加载所需要的模块

# 加载需要用到的模块
import scikitplot as skplt

import sklearn
from sklearn.datasets import load_digits, load_boston, load_breast_cancer
from sklearn.model_selection import train_test_split

from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor, GradientBoostingClassifier, ExtraTreesClassifier
from sklearn.linear_model import LinearRegression, LogisticRegression
from sklearn.cluster import KMeans
from sklearn.decomposition import PCA

import matplotlib.pyplot as plt

import sys

print("Scikit Plot Version : ", skplt.__version__)
print("Scikit Learn Version : ", sklearn.__version__)
print("Python Version : ", sys.version)

2加载数据集

2.1手写数据集

digits = load_digits()
X_digits, Y_digits = digits.data, digits.target

print("Digits Dataset Size : ", X_digits.shape, Y_digits.shape)

X_digits_train, X_digits_test, Y_digits_train, Y_digits_test = train_test_split(X_digits, Y_digits,
                                                                                train_size=0.8,
                                                                                stratify=Y_digits,
                                                                                random_state=1)

print("Digits Train/Test Sizes : ",X_digits_train.shape, X_digits_test.shape, Y_digits_train.shape, Y_digits_test.shape)

Digits Dataset Size :  (1797, 64) (1797,) Digits Train/Test Sizes :  (1437, 64) (360, 64) (1437,) (360,)

2.2、肿瘤数据集

cancer = load_breast_cancer()
X_cancer, Y_cancer = cancer.data, cancer.target

print("Feautre Names : ", cancer.feature_names)
print("Cancer Dataset Size : ", X_cancer.shape, Y_cancer.shape)
X_cancer_train, X_cancer_test, Y_cancer_train, Y_cancer_test = train_test_split(X_cancer, Y_cancer,
                                                                                train_size=0.8,
                                                                                stratify=Y_cancer,
                                                                                random_state=1)

print("Cancer Train/Test Sizes : ",X_cancer_train.shape, X_cancer_test.shape, Y_cancer_train.shape, Y_cancer_test.shape)

Feautre Names :  ['mean radius' 'mean texture' 'mean perimeter' 'mean area'
 'mean smoothness' 'mean compactness' 'mean concavity'
 'mean concave points' 'mean symmetry' 'mean fractal dimension'
 'radius error' 'texture error' 'perimeter error' 'area error'
 'smoothness error' 'compactness error' 'concavity error'
 'concave points error' 'symmetry error' 'fractal dimension error'
 'worst radius' 'worst texture' 'worst perimeter' 'worst area'
 'worst smoothness' 'worst compactness' 'worst concavity'
 'worst concave points' 'worst symmetry' 'worst fractal dimension']
Cancer Dataset Size :  (569, 30) (569,)
Cancer Train/Test Sizes :  (455, 30) (114, 30) (455,) (114,)

2.3波斯顿房价数据集

boston = load_boston()
X_boston, Y_boston = boston.data, boston.target

print("Boston Dataset Size : ", X_boston.shape, Y_boston.shape)

print("Boston Dataset Features : ", boston.feature_names)
X_boston_train, X_boston_test, Y_boston_train, Y_boston_test = train_test_split(X_boston, Y_boston,
                                                                                train_size=0.8,
                                                                                random_state=1)

print("Boston Train/Test Sizes : ",X_boston_train.shape, X_boston_test.shape, Y_boston_train.shape, Y_boston_test.shape)

Boston Dataset Size :  (506, 13) (506,)
Boston Dataset Features :  ['CRIM' 'ZN' 'INDUS' 'CHAS' 'NOX' 'RM' 'AGE' 'DIS' 'RAD' 'TAX' 'PTRATIO'
 'B' 'LSTAT']
Boston Train/Test Sizes :  (404, 13) (102, 13) (404,) (102,)

3性能可视化

3.1交叉验证绘制

skplt.estimators.plot_learning_curve(LogisticRegression(), X_digits, Y_digits,
                                     cv=7, shuffle=True, scoring="accuracy",
                                     n_jobs=-1, figsize=(6,4), title_fontsize="large", text_fontsize="large",
                                     title="Digits Classification Learning Curve")
plt.show()

skplt.estimators.plot_learning_curve(LinearRegression(), X_boston, Y_boston,
                                     cv=7, shuffle=True, scoring="r2", n_jobs=-1,
                                     figsize=(6,4), title_fontsize="large", text_fontsize="large",
                                     title="Boston Regression Learning Curve ");
plt.show() 

3.2 重要性特征绘制

rf_reg = RandomForestRegressor()
rf_reg.fit(X_boston_train, Y_boston_train)
print(rf_reg.score(X_boston_test, Y_boston_test))
gb_classif = GradientBoostingClassifier()
gb_classif.fit(X_cancer_train, Y_cancer_train)
print(gb_classif.score(X_cancer_test, Y_cancer_test))

fig = plt.figure(figsize=(15,6))

ax1 = fig.add_subplot(121)
skplt.estimators.plot_feature_importances(rf_reg, feature_names=boston.feature_names,
                                         title="Random Forest Regressor Feature Importance",
                                         x_tick_rotation=90, order="ascending",
                                         ax=ax1);

ax2 = fig.add_subplot(122)
skplt.estimators.plot_feature_importances(gb_classif, feature_names=cancer.feature_names,
                                         title="Gradient Boosting Classifier Feature Importance",
                                         x_tick_rotation=90,
                                         ax=ax2);

plt.tight_layout()
plt.show()

0.9096714634570764
0.9649122807017544

4 机器学习度量(metrics)

4.1 混淆矩阵(Confusion Matrix)

log_reg = LogisticRegression()
log_reg.fit(X_digits_train, Y_digits_train)
log_reg.score(X_digits_test, Y_digits_test)
Y_test_pred = log_reg.predict(X_digits_test)

fig = plt.figure(figsize=(15,6))

ax1 = fig.add_subplot(121)
skplt.metrics.plot_confusion_matrix(Y_digits_test, Y_test_pred,
                                    title="Confusion Matrix",
                                    cmap="Oranges",
                                    ax=ax1)

ax2 = fig.add_subplot(122)
skplt.metrics.plot_confusion_matrix(Y_digits_test, Y_test_pred,
                                    normalize=True,
                                    title="Confusion Matrix",
                                    cmap="Purples",
                                    ax=ax2);
plt.show()

4.2 ROC、AUC曲线

Y_test_probs = log_reg.predict_proba(X_digits_test)

skplt.metrics.plot_roc_curve(Y_digits_test, Y_test_probs,
                       title="Digits ROC Curve", figsize=(12,6))
plt.show()

4.3 PR曲线

skplt.metrics.plot_precision_recall_curve(Y_digits_test, Y_test_probs,
                       title="Digits Precision-Recall Curve", figsize=(12,6))
plt.show()

4.4 轮廓分析（Silhouette Analysis）

kmeans = KMeans(n_clusters=10, random_state=1)
kmeans.fit(X_digits_train, Y_digits_train)
cluster_labels = kmeans.predict(X_digits_test)
skplt.metrics.plot_silhouette(X_digits_test, cluster_labels,
                              figsize=(8,6))
plt.show()

4.5 可靠性曲线（Calibration Curve ，Reliability Curves）

lr_probas = LogisticRegression().fit(X_cancer_train, Y_cancer_train).predict_proba(X_cancer_test)
rf_probas = RandomForestClassifier().fit(X_cancer_train, Y_cancer_train).predict_proba(X_cancer_test)
gb_probas = GradientBoostingClassifier().fit(X_cancer_train, Y_cancer_train).predict_proba(X_cancer_test)
et_scores = ExtraTreesClassifier().fit(X_cancer_train, Y_cancer_train).predict_proba(X_cancer_test)

probas_list = [lr_probas, rf_probas, gb_probas, et_scores]
clf_names = ['Logistic Regression', 'Random Forest', 'Gradient Boosting', 'Extra Trees Classifier']
skplt.metrics.plot_calibration_curve(Y_cancer_test,
                                     probas_list,
                                     clf_names, n_bins=15,
                                     figsize=(12,6)
                                     )
plt.show()

4.6 KS检验

f = RandomForestClassifier()
rf.fit(X_cancer_train, Y_cancer_train)
Y_cancer_probas = rf.predict_proba(X_cancer_test)

skplt.metrics.plot_ks_statistic(Y_cancer_test, Y_cancer_probas, figsize=(10,6))
plt.show()

4.7 累积收益曲线

skplt.metrics.plot_cumulative_gain(Y_cancer_test, Y_cancer_probas, figsize=(10,6))
plt.show()

4.8 Lift 曲线

kplt.metrics.plot_lift_curve(Y_cancer_test, Y_cancer_probas, figsize=(10,6))
plt.show()

5 聚类方法

5.1 手肘法（Elbow Method）

skplt.cluster.plot_elbow_curve(KMeans(random_state=1),
                               X_digits,
                               cluster_ranges=range(2, 20),
                               figsize=(8,6))
plt.show()

6 降维方法

6.1 PCA

pca = PCA(random_state=1)
pca.fit(X_digits)

skplt.decomposition.plot_pca_component_variance(pca, figsize=(8,6))
plt.show()

6.2 2-D Projection

skplt.decomposition.plot_pca_2d_projection(pca, X_digits, Y_digits,
                                           figsize=(10,10),
                                           cmap="tab10")
plt.show()

7 可视化决策树的方式

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn.datasets import load_iris
from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import train_test_split
from sklearn import tree
import graphviz
# 加载数据集
dataset = load_iris() 
# 转换成.DataFrame形式
df = pd.DataFrame(dataset.data, columns = dataset.feature_names)
# 添加品种列
df['Species'] = dataset.target

# 用数值替代品种名作为标签
target = np.unique(dataset.target)
target_names = np.unique(dataset.target_names)
targets = dict(zip(target, target_names))
df['Species'] = df['Species'].replace(targets)

# 提取数据和标签
X = df.drop(columns="Species")
y = df["Species"]
feature_names = X.columns
labels = y.unique()

X_train, X_test, y_train, y_test = train_test_split(X,y,
                                                 test_size = 0.4,
                                                 random_state = 42)
model = DecisionTreeClassifier(max_depth =3, random_state = 42)
model.fit(X_train, y_train)     

DecisionTreeClassifier(max_depth=3, random_state=42)

7.1文字表示

# 以文字形式输出树     
text_representation = tree.export_text(model)
print(text_representation)

|--- feature_2 <= 2.45
|   |--- class: setosa
|--- feature_2 >  2.45
|   |--- feature_3 <= 1.75
|   |   |--- feature_2 <= 5.35
|   |   |   |--- class: versicolor
|   |   |--- feature_2 >  5.35
|   |   |   |--- class: virginica
|   |--- feature_3 >  1.75
|   |   |--- feature_2 <= 4.85
|   |   |   |--- class: virginica
|   |   |--- feature_2 >  4.85
|   |   |   |--- class: virginica

7.2图片形式（plot_tree函数）

# 用图片画出
plt.figure(figsize=(30,10), facecolor ='g') #
a = tree.plot_tree(model,
                   feature_names = feature_names,
                   class_names = labels,
                   rounded = True,
                   filled = True,
                   fontsize=14)
plt.show() 

7.3 graphviz模块

该模块需要安装graphviz，推荐在ubuntu上sudo apt install graphviz或者Mac上brew install graphviz

dot_data = tree.export_graphviz(model, out_file=None, 
                                feature_names=dataset.feature_names,  
                                class_names=dataset.target_names,
                                filled=True)

# Draw graph
graph = graphviz.Source(dot_data, format="png") 
graph

7.4 sns.heatmap可视化报告

from sklearn.metrics import classification_report
import numpy as np
import seaborn as sns
import pandas as pd
y_pred = model.predict(X_test)
target_names = np.unique(dataset.target_names)
clf_report = classification_report(y_test,
                                   y_pred,
                                   labels=labels,
                                   target_names=target_names,
                                   output_dict=True)
plt.figure(figsize=(8,6))
sns.heatmap(pd.DataFrame(clf_report).iloc[:-1, :].T, annot=True)
plt.show()

8 总结

Scikit-plot模块与sklearn完美兼容,主要提供的可视化指标图归为两类，分别是：聚类和分类任务。当然了对于数据降维处理等方面也有提供基础的方法，其提供的主要方法都是我们使用频度很高的，对于每一位机器学习从业者来说都可以节省很多不必要的时间开销，从而大幅提升工作效率。