ハードマージンSVM-Breast Cancer Wisconsin (Diagnostic) Data Set-

やりたいこと

サポートベクトルマシンを作ってみる。
やってみた感想としては、libsvmのまま使ったほうが使いやすいかも？

コード

'''
Created on 2019/05/15

@author: tatsunidas
'''
import numpy as np
import matplotlib.pyplot as plt
from matplotlib.colors import ListedColormap
from sklearn import svm
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
import pandas as pd
import seaborn as sns # used for plot interactive graph. 
#データの読み込み
from sklearn.datasets import load_breast_cancer
dataset = load_breast_cancer()
df = pd.DataFrame(data=dataset.data, columns=dataset.feature_names)

# トレーニングデータとテストデータに分割
# 乱数を制御するパラメータ random_state は None にすると毎回異なるデータを生成する
X, X_test, y, y_test = train_test_split(dataset.data, dataset.target, test_size=0.2, random_state=None)
print(X.shape)

# データの標準化処理
sc = StandardScaler()
sc.fit(X)
X = sc.transform(X)
X_test = sc.transform(X_test)

# fit the model
h = .02  # step size in the mesh
# we create an instance of SVM and fit out data. We do not scale our
# data since we want to plot the support vectors
svc = svm.SVC(kernel='linear', C=1.).fit(X, y)#One-vs-One
lin_svc = svm.LinearSVC(C=1., tol=2e-03).fit(X, y)#One-vs-All
rbf_svc = svm.NuSVC(kernel='rbf', gamma='auto').fit(X, y)
# polyのみdegreeをオプションで使える。チューニングは難しい
poly_svc = svm.NuSVC(kernel='poly', gamma='auto', degree=3,decision_function_shape='ovr', nu=0.05).fit(X, y)

#show accuracy
print('score :: 1 svc, 2 lin_svc, 3 rbf_svc, 4 poly_svc')
for j, clf in enumerate((svc, lin_svc, rbf_svc, poly_svc)):
    print(str(j+1))
    print('正解率(train):{:.3f}'.format(clf.score(X,y)))
    print('正解率(test):{:.3f}'.format(clf.score(X_test,y_test)))

'''
graphを作ってみる
'''
#相関を見てみる
df = pd.DataFrame(data=dataset.data, columns=dataset.feature_names)
corr = df.corr() # .corr is used to find corelation
f,ax = plt.subplots(figsize=(20, 20))
sns.heatmap(corr, cbar = True,  square = True, annot = True, fmt= '.1f', 
            xticklabels= True, yticklabels= True
            ,cmap="coolwarm", linewidths=.5, ax=ax);
plt.show()

#最初の2つの特徴を使って境界を描画してみる
# create a mesh to plot in
x_min, x_max = X[:, 0].min() - 1, X[:, 0].max() + 1
y_min, y_max = X[:, 1].min() - 1, X[:, 1].max() + 1
xx, yy = np.meshgrid(np.arange(x_min, x_max, h),
                     np.arange(y_min, y_max, h))

# title for the plots
titles = ['SVC with linear kernel',
          'LinearSVC (linear kernel)',
          'SVC with RBF kernel',
          'SVC with polynomial kernel']
plt.figure(figsize=(10,12))
for j, clf in enumerate((svc, lin_svc, rbf_svc, poly_svc)):
    # Plot the decision boundary by assigning a color to each point in the mesh 
    plt.subplot(2, 2, j + 1)
    plt.subplots_adjust(wspace=0.4, hspace=0.4)

    pre_pred = np.array([xx.ravel(), yy.ravel()] + [np.repeat(0, xx.ravel().size) for _ in range(28)]).T
    print(pre_pred.shape)
    pred = clf.predict(pre_pred)
    Z = pred.reshape(xx.shape)
    decf = clf.decision_function(pre_pred).reshape(xx.shape)    

    # Put the result into a color plot
    # decision func
    plt.contourf(xx, yy, decf,alpha=1.0, cmap="RdYlGn", levels=np.linspace(decf.min(), decf.max(), 100))
    # boundary (これを使うときは一行上の plt.contourf(xx, yy, decf...をコメントアウトして)
#     plt.contourf(xx, yy, Z, cmap=plt.cm.get_cmap('RdBu_r'), alpha=0.6)

# Ploting  the training points
    plt.scatter(X[:, 0], X[:, 1], c=y, cmap=plt.cm.get_cmap('RdBu_r'))
    plt.xlabel(dataset.feature_names[0],size=20)
    plt.ylabel(dataset.feature_names[1],size=20)
    plt.xlim(xx.min(), xx.max())
    plt.ylim(yy.min(), yy.max())
    plt.xticks(())
    plt.yticks(())
    plt.title(titles[j],size=20);

plt.show()

# 5/17 added
# ROC curve(ロックカーブ)
titles = ['SVC with linear kernel',
#           'LinearSVC (linear kernel)',　# proba関数がない。予測を取得したいだけなのでpredict関数でもOKと思う。
          'SVC with RBF kernel',
          'SVC with polynomial kernel']
plt.figure(figsize=(10,12))
for j, clf in enumerate((svc, rbf_svc, poly_svc)):# exclude lin_svc
    # Plot each ROC curve
    plt.subplot(2, 2, j + 1)
    plt.subplots_adjust(wspace=0.4, hspace=0.4)
    #予測確率を取得
    pred = clf.predict_proba(X_test)[:,1]
    #偽陽性率と真陽性率の算出
    fpr, tpr, thretholds = roc_curve(y_test,pred) 
    # aucの算出
#     auc = auc(fpr,tpr)
    auc = roc_auc_score(y_test, pred)
    # Ploting  ROC
    plt.plot(fpr, tpr,label='ROC curve (auc = %.2f)'%(auc), marker='o', color='red')
    plt.plot([0,1],[0,1], color = 'black',linestyle='--')

    plt.xlabel('false positive rate')
    plt.ylabel('true positive rate')
    plt.xlim(-0.05, 1.0)
    plt.ylim(0.0, 1.05)
    plt.title('receiver operating characteristic'+'\n'+ titles[j]);
    plt.legend()
    plt.grid()

plt.show()

結果

(455, 30)#形状の確認用
score :: 1 svc, 2 lin_svc, 3 rbf_svc, 4 poly_svc
1
正解率(train):0.987
正解率(test):0.991
2
正解率(train):0.989
正解率(test):0.982
3
正解率(train):0.947
正解率(test):0.965
4
正解率(train):0.998
正解率(test):0.974
(175616, 30)
(175616, 30)
(175616, 30)
(175616, 30)

途中で描画するグラフ

データセット内の特徴相関図

最初の2つの特徴を使って描画した境界面

（polyはチューニングが難しい。）

ROC curve

参考文献やURL

https://qiita.com/kazuki_hayakawa/items/18b7017da9a6f73eba77
https://scikit-learn.org/stable/modules/generated/sklearn.svm.NuSVC.html
https://www.kaggle.com/rcfreitas/python-ml-breast-cancer-diagnostic-data-set
https://ameblo.jp/cognitive-solution/entry-12290094948.html
https://scikit-learn.org/stable/auto_examples/svm/plot_svm_nonlinear.html#sphx-glr-auto-examples-svm-plot-svm-nonlinear-py
https://www.kaggle.com/selener/breast-cancer-diagnosis/notebook
https://stackoverflow.com/questions/45384185/what-is-the-difference-between-linearsvc-and-svckernel-linear/45390526
https://www.kaggle.com/sachin1512/breast-cancer-dataset/downloads/breast-cancer-dataset.zip/1
https://archive.ics.uci.edu/ml/datasets/breast+cancer+wisconsin+(original)