Get Confusion Matrix From a Keras Multiclass Model [duplicate]

I am building a multiclass model with Keras.

model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
model.fit(X_train, y_train, batch_size=batch_size, epochs=epochs, verbose=1, callbacks=[checkpoint], validation_data=(X_test, y_test)) # starts training

Here is how my test data looks like (it's text data).

X_test
Out[25]:
array([[621, 139, 549, ..., 0, 0, 0], [621, 139, 543, ..., 0, 0, 0]])
y_test
Out[26]:
array([[0, 0, 1], [0, 1, 0]])

After generating predictions...

predictions = model.predict(X_test)
predictions
Out[27]:
array([[ 0.29071924, 0.2483743 , 0.46090645], [ 0.29566404, 0.45295066, 0.25138539]], dtype=float32)

I did the following to get the confusion matrix.

y_pred = (predictions > 0.5)
confusion_matrix(y_test, y_pred)
Traceback (most recent call last): File "<ipython-input-38-430e012b2078>", line 1, in <module> confusion_matrix(y_test, y_pred) File "/Users/abrahammathew/anaconda3/lib/python3.6/site-packages/sklearn/metrics/classification.py", line 252, in confusion_matrix raise ValueError("%s is not supported" % y_type)
ValueError: multilabel-indicator is not supported

However, I am getting the above error.

How can I get a confusion matrix when doing a multiclass neural network in Keras?

1 Answer

Your input to confusion_matrix must be an array of int not one hot encodings.

matrix = metrics.confusion_matrix(y_test.argmax(axis=1), y_pred.argmax(axis=1))