ValueError: Classification metrics can't handle a mix of binary and continuous targets

Question

y_test.shape

output: (40,)

predict.shape

output: (40,1)

confusion_matrix(y_test, predict)

Error: ValueError: Classification metrics can't handle a mix of binary and continuous targets

I am applying Confusion Metrix at the prediction of CNN model on textual data, but an error message got that confusion matrix cannot handle mix of binary and continuous data

Answer 1

This sometimes happens when a classifier produces quasi-probabilistic estimates, such as when a crossentropy loss is used in a neural.network.

import numpy as np

y_test = np.array([0, 1, 0])
y_pred = np.array([[0.1], [0.7], [0.2]])

print(y_test.shape, y_pred.shape)
# (3,)    (3, 1)

This is solved by thresholding the predictions. 0.5 is a common choice:

from sklearn.metrics import confusion_matrix

confusion_matrix(
  y_test,
  (y_pred > 0.5),
)
# array([[2, 0],
#        [0, 1]])