sklearn中多类别分类情况下的平均折线ROC曲线

Question

我有一个共3类的多分类问题。 我使用LinearDiscriminantAnalysis进行分类，我想在KFolds上绘制平均值 ROC（k = 5）。

对于二进制分类案例，我能够做到这一点，但我找不到找到使之适用于多分类案例的方法。

以下是我针对 binary 案例的代码：

import matplotlib.pyplot as plt
import numpy as np
from scipy import interp
from sklearn.datasets import make_classification
from sklearn.cross_validation import KFold
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.model_selection import cross_val_predict
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.model_selection import KFold
from sklearn.metrics import roc_curve, auc
from itertools import cycle
from scipy import interp
from sklearn.preprocessing import label_binarize
from sklearn.multiclass import OneVsRestClassifier
plt.style.use('ggplot')


X, y = make_classification(n_samples=500, random_state=100, flip_y=0.3)

kf = KFold(n_splits = 5, shuffle = True, random_state= 0)
clf = LinearDiscriminantAnalysis()
pipe= Pipeline([('scaler', StandardScaler()), ('clf', clf)])

tprs = []
aucs = []
base_fpr = np.linspace(0, 1, 101)
colors = ['darksalmon', 'gold', 'royalblue', 'mediumseagreen', 'violet']

for i, (train, test) in enumerate(kf.split(X,y)):
    model = pipe.fit(X[train], y[train])
    y_score = model.predict_proba(X[test])
    fpr, tpr, _ = roc_curve(y[test], y_score[:, 1])
    roc_auc = auc(fpr, tpr)
    aucs.append(roc_auc)
    #plt.plot(fpr, tpr, lw=1, alpha=0.6, label='ROC fold %d (AUC = %0.2f)' % (i, roc_auc), c = colors[i])
    tpr = interp(base_fpr, fpr, tpr)
    tpr[0] = 0.0
    tprs.append(tpr)

tprs = np.array(tprs)
mean_tprs = tprs.mean(axis=0)
std = tprs.std(axis=0)

mean_auc = auc(base_fpr, mean_tprs)
std_auc = np.std(aucs)

tprs_upper = np.minimum(mean_tprs + std, 1)
tprs_lower = mean_tprs - std

plt.figure(figsize=(12, 8))
plt.plot(base_fpr, mean_tprs, 'b', alpha = 0.8, label=r'Mean ROC (AUC = %0.2f $\pm$ %0.2f)' % (mean_auc, std_auc),)
plt.fill_between(base_fpr, tprs_lower, tprs_upper, color = 'blue', alpha = 0.2)
plt.plot([0, 1], [0, 1], linestyle = '--', lw = 2, color = 'r', label = 'Luck', alpha= 0.8)
plt.xlim([-0.01, 1.01])
plt.ylim([-0.01, 1.01])
plt.ylabel('True Positive Rate')
plt.xlabel('False Positive Rate')
plt.legend(loc="lower right")
plt.title('Receiver operating characteristic (ROC) curve')
#plt.axes().set_aspect('equal', 'datalim')
plt.show()

编辑1：

我尝试使用OneVsRestClassifier使它适用于多类案例：

import matplotlib.pyplot as plt
import numpy as np
from scipy import interp
from sklearn.datasets import make_classification
from sklearn.cross_validation import KFold
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.model_selection import cross_val_predict
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.model_selection import KFold
from sklearn.metrics import roc_curve, auc
from itertools import cycle
from scipy import interp
from sklearn.multiclass import OneVsRestClassifier
from sklearn.model_selection import StratifiedKFold
from sklearn.preprocessing import label_binarize
plt.style.use('ggplot')
plt.figure(figsize=(12, 8))


X, y = make_classification(n_samples=500, random_state=100, n_classes=3,n_clusters_per_class=1, flip_y=0.3)

kf = KFold(n_splits = 5, shuffle = True, random_state= 0)
clf = OneVsRestClassifier(LinearDiscriminantAnalysis())
pipe= Pipeline([('scaler', StandardScaler()), ('clf', clf)])
classes = np.unique(y)
y_true = label_binarize(y, classes=classes)
n_classes = y_true.shape[1]

base_fpr = np.linspace(0, 1, 101)
colors = ['darksalmon', 'gold', 'royalblue', 'mediumseagreen', 'violet']
fpr = dict()
tpr = dict()
roc_auc = dict()

fff=[]
ttt=[]
aucc=[]
# Fit the model for each fold
for i, (train, test) in enumerate(kf.split(X,y)):
    model = pipe.fit(X[train], y[train])
    y_score = model.predict_proba(X[test])

    # Compute ROC curve and ROC area for each class PER FOLD
    for j in range(n_classes):
        fpr[j], tpr[j], _ = roc_curve(y_true[test][:, j], y_score[:, j])
        roc_auc[j] = auc(fpr[j], tpr[j])

    # First aggregate all false positive rates per classe for each fold
    all_fpr = np.unique(np.concatenate([fpr[j] for j in range(n_classes)]))

    # Then interpolate all ROC curves at this points
    mean_tpr = np.zeros_like(all_fpr)
    for j in range(n_classes):
        mean_tpr += interp(all_fpr, fpr[j], tpr[j])

    # Finally average it and compute AUC for EACH FOLD
    mean_tpr /= n_classes

    fpr["macro"] = all_fpr
    tpr["macro"] = mean_tpr
    roc_auc["macro"] = auc(fpr["macro"], tpr["macro"])

    fff.append(all_fpr)
    ttt.append(mean_tpr)
    aucc.append(roc_auc["macro"])

# Compute average across Folds
fff = np.array(fff)
ttt = np.array(ttt)
aucc = np.array(aucc)

all_fpr_folds = np.unique(np.concatenate([fff[j] for j in range(kf.get_n_splits())]))

# Then interpolate all ROC curves at this points
mean_tpr_folds = np.zeros_like(all_fpr_folds)
for j in range(kf.get_n_splits()):
    mean_tpr_folds += interp(all_fpr_folds, fff[j], ttt[j])

# Finally average it and compute AUC
mean_tpr_folds /= float(kf.get_n_splits())


mean_mean_tpr_folds= mean_tpr_folds.mean(axis = 0)
std = mean_tpr_folds.std(axis=0)

tprs_upper = np.minimum(mean_mean_tpr_folds + std, 1)
tprs_lower = mean_mean_tpr_folds - std

plt.plot(all_fpr_folds, mean_tpr_folds, 'b', alpha = 0.8, label=r'Mean ROC (AUC = %0.2f $\pm$ %0.2f)' % (aucc.mean(), aucc.std()),)
plt.fill_between(all_fpr_folds, tprs_lower, tprs_upper, color = 'blue', alpha = 0.2)
plt.plot([0, 1], [0, 1], linestyle = '--', lw = 2, color = 'r', label = 'Luck', alpha= 0.8)
plt.xlim([-0.01, 1.01])
plt.ylim([-0.01, 1.01])
plt.ylabel('True Positive Rate')
plt.xlabel('False Positive Rate')
plt.legend(loc="lower right")
plt.title('Receiver operating characteristic (ROC) curve')
#plt.axes().set_aspect('equal', 'datalim')
plt.show()

我在这里丢失了一些东西...