pythonmachine-learningtheanoconvolutionkeras

How to find wrong prediction cases in test set (CNNs using Keras)

I'm using MNIST example with 60000 training image and 10000 testing image. How do I find which of the 10000 testing image that has an incorrect classification/prediction?

Solution

  • Simply use model.predict_classes() and compare the output with true labes. i.e:

    incorrects = np.nonzero(model.predict_class(X_test).reshape((-1,)) != y_test)

    to get indices of incorrect predictions