结合sklearn的可视化工具Yellowbrick：超参与行为的可视化带来更优秀的实现

Yellowbrick 是一套名为「Visualizers」的视觉诊断工具，它扩展了 Scikit-Learn API 以允许我们监督模型的选择过程。简而言之，Yellowbrick 将 Scikit-Learn 与 Matplotlib 结合在一起，并以传统 Scikit-Learn 的方式对模型进行可视化。

• 可视化器
  可视化器（Visualizers）是一种从数据中学习的估计器，其主要目标是创建可理解模型选择过程的可视化。在 Scikit-Learn 的术语中，它们类似于转换器（transformer），其在可视化数据空间或包装模型估计器上类似「ModelCV」（例如 RidgeCV 和 LassoCV）方法的过程。Yellowbrick 的主要目标是创建一个类似于 Scikit-Learn 的 API，其中一些流行的可视化器包括：

  特征可视化

  • Rank Features：单个或成对特征排序以检测关系
  • Radial Visualization：围绕圆形图分离实例
  • PCA Projection：基于主成分分析映射实例
  • Manifold Visualization：通过流形学习实现高维可视化
  • Feature Importances：基于模型性能对特征进行排序
  • Recursive Feature Elimination：按重要性搜索最佳特征子集
  • Scatter and Joint Plots：通过特征选择直接进行数据可视化

  分类可视化

  • Class Balance：了解类别分布如何影响模型
  • Class Prediction Error：展示分类的误差与主要来源
  • Classification Report：可视化精度、召回率和 F1 分数的表征
  • ROC/AUC Curves：受试者工作曲线和曲线下面积
  • Confusion Matrices：类别决策制定的视觉描述
  • Discrimination Threshold：搜索最佳分离二元类别的阈值

  回归可视化

  • Prediction Error Plots：沿着目标域寻找模型崩溃的原因
  • Residuals Plot：以残差的方式展示训练和测试数据中的差异
  • Alpha Selection：展示 alpha 的选择如何影响正则化

  聚类可视化

  • K-Elbow Plot：使用肘法（elbow method）和多个指标来选择 k
  • Silhouette Plot：通过可视化轮廓系数值来选择 k

  模型选择可视化

  • Validation Curve：对模型的单个超参数进行调整
  • Learning Curve：展示模型是否能从更多的数据或更低的复杂性中受益

  文本可视化

  • Term Frequency：可视化语料库中词项的频率分布
  • t-SNE Corpus Visualization：使用随机近邻嵌入来投影文档

  实例

#特征之间协方差可视化
from yellowbrick.features import Rank2D
from sklearn.datasets import load_iris
data=load_iris()
visualizer = Rank2D(features=data['feature_names'], algorithm='covariance')
visualizer.fit(data['data'], data['target']) # Fit the data to the visualizer
visualizer.transform(data['data']) # Transform the data
visualizer.poof() # Draw/show/poof the data

#梯度提升树中特征重要性可视化
import matplotlib.pyplot as plt
from sklearn.ensemble import GradientBoostingClassifier
from yellowbrick.features import FeatureImportances
from sklearn.datasets import load_iris
data=load_iris()
fig = plt.figure()
ax = fig.add_subplot()
viz = FeatureImportances(GradientBoostingClassifier(), relative=False)
viz.fit(data['data'],data['target']) # Fit the data to the visualizer
viz.poof() # Draw/show/poof the data

#线性支持向量机ROC曲线可视化
from sklearn.svm import LinearSVC
from yellowbrick.classifier import ROCAUC

model = LinearSVC()
model.fit(data['data'],data['target'])
visualizer = ROCAUC(model)
visualizer.score(data['data'],data['target'])
visualizer.poof()

#主成分分析二维降维可视化
from yellowbrick.features.pca import PCADecomposition

visualizer = PCADecomposition(scale=True, center=False, color="g", proj_dim=2)
visualizer.fit_transform(data['data'],data['target'])
visualizer.poof()

#线性支持向量机准确率、召回率、f1-score可视化
from sklearn.svm import LinearSVC
from yellowbrick.classifier import ClassificationReport
from sklearn.model_selection import train_test_split
model = LinearSVC()
X_train, X_test, y_train, y_test = train_test_split(data['data'],data['target'], test_size=0.2)
visualizer = ClassificationReport(model, classes=data['target_names'])
visualizer.fit(X_train, y_train)  # Fit the visualizer and the model
visualizer.score(X_test, y_test)  # Evaluate the model on the test data
g = visualizer.poof()             # Draw/show/poof the data

#alpha 的选择如何影响正则化可视化
import numpy as np

from sklearn.linear_model import LassoCV
from yellowbrick.regressor import AlphaSelection

# Create a list of alphas to cross-validate against
alphas = np.logspace(-10, 1, 400)#以10为底对数，-10到1分成400份

# Instantiate the linear model and visualizer
model = LassoCV(alphas=alphas)
visualizer = AlphaSelection(model)

visualizer.fit(data['data'],data['target'])
g = visualizer.poof()

#肘部法则选择最佳聚类的k
from sklearn.cluster import MiniBatchKMeans

from yellowbrick.cluster import KElbowVisualizer

# Instantiate the clustering model and visualizer
visualizer = KElbowVisualizer(MiniBatchKMeans(), k=(4,12))

visualizer.fit(data["data"]) # Fit the training data to the visualizer
visualizer.poof() # Draw/show/poof the data

#训练集数量对模型表现可视化
import numpy as np

from sklearn.naive_bayes import MultinomialNB
from sklearn.model_selection import StratifiedKFold
from yellowbrick.model_selection import LearningCurve

# Create the learning curve visualizer
cv = StratifiedKFold(12)#k折交叉切分
sizes = np.linspace(0.3, 1.0, 10)

viz = LearningCurve(
    MultinomialNB(), cv=cv, train_sizes=sizes,
    scoring='f1_weighted', n_jobs=4
)

# Fit and poof the visualizer
viz.fit(data['data'],data['target'])
viz.poof()