Assuming that I have the following:

x_train
y_train
x_test
y_test

And there is 7 labels total (0 through 6), I am trying to create a ROC curve for each individual label type of the testing data. I already looked at this post,https://stackoverflow.com/questions/65236646/valueerror-multilabel-indicator-format-is-not-supported-for-roc-curve-sklea and tried to create a code that would work, however I am having bad luck.

I ran the model for the datasets, which is a CNN model:

history_model = model.fit(
    x_train, y_train,
    epochs=1000,
    batch_size = 16,
    validation_data=(x_test, y_test),
    verbose=2)

In order to visualize the data, I am trying to make a ROC curve, which can help me understand the TP rate and FP rate. Now, I tried the following to make a ROC curve...

y_pred = model.predict(x_test)
fpr, tpr, threshold = metrics.roc_curve(y_test, y_pred)
roc_auc = metrics.auc(fpr, tpr)

However, I am getting the following error: ValueError: multilabel-indicator format is not supported

What can be done to get around this issue? I'm not too sure what I can do. Please help.

CodePudding user response：

The roc_curve and auc functions only work on 1-d arrays. In your case, you must loop for each label.

fpr_list = []
tpr_list = []
threshold_list = []
roc_auc_list = []

for i in range(7): # you can make this more general
    fpr, tpr, threshold = metrics.roc_curve(y_test[:, i], y_pred[:, i])
    roc_auc = metrics.auc(fpr, tpr)
    
    fpr_list.append(fpr)
    tpr_list.append(tpr)
    threshold_list.append(threshold)
    roc_auc_list.append(roc_auc)