Why am i getting different roc_auc_scores?

Question

I am working on a binary classification problem and using the following code to find ROC_AUC score(at each fold) using 5-fold cross-validation.

cv = KFold(n_splits=5,shuffle=True,random_state=41)
classifier=RandomForestClassifier()
tprs = []
aucs = []
mean_fpr = np.linspace(0, 1, 100)
for i,(train, test) in enumerate (cv.split(X, y)):
    clf=classifier.fit(X[train],y[train])
    prediction=clf.predict_proba(X[test])[:,1]
    fpr,tpr,t=roc_curve(y[test],prediction)
    tprs.append(np.interp(mean_fpr,fpr,tpr))
    roc_auc=auc(fpr,tpr)
    aucs.append(roc_auc)
    plt.plot(fpr,tpr,lw=2,alpha=1,label='Fold %d (AUC=%0.4f)'%(i+1,roc_auc))
plt.plot([0,1], [0,1], linestyle='--', lw=2, color='k',
        label='Chance', alpha=1)
mean_tpr = np.mean(tprs, axis=0)
mean_auc = auc(mean_fpr, mean_tpr)
std_auc = np.std(aucs)
plt.plot(mean_fpr, mean_tpr, color='b',
      label=r'Mean ROC (AUC = %0.4f $\pm$ %0.4f)' % (mean_auc, std_auc),
       lw=2, alpha=1)
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.title('ROC')
plt.legend(loc="lower right")
plt.show()

In addition to ROC_AUC scores, I also want to calculate other performance metrics like accuracy, precision, recall at each fold. So for that I used the following code, but I am getting different scores for ROC_AUC here compared to what I obtained above.

from sklearn.metrics import make_scorer,accuracy_score,precision_score,recall_score,roc_auc_score
scoring={'accuracy': make_scorer(accuracy_score),
         'recall': make_scorer(recall_score),
        'precision': make_scorer(precision_score),
        'roc_auc_score':make_scorer(roc_auc_score)}

results=model_selection.cross_validate(estimator=classifier,X=X,y=y,cv=cv,scoring=scoring)

Can anyone point out where I am going wrong? Also how to calculate other metrics at each fold?

Answer 1

Have a look at the make_scorer function:

metrics.make_scorer(score_func, *, greater_is_better=True, needs_proba=False, needs_threshold=False, **kwargs)

By default, the argument 'needs_proba' is set to False. However, in case of roc_auc_score it should be set to True. If it is set to False, the ROC_AUC score will be calculated by using only one confidence threshold instead of multiples and therefore the two results differ.