Keras: .predict returns percentages instead of classes

I am building a model with 3 classes: [0,1,2]
After training, the .predict function returns a list of percentages instead.
I was checking the keras documentation but could not figure out, what I did wrong.
.predict_classes is not working anymore, and I did not have this problem with previous classifiers. I already tried different activation functions (relu, sigmoid etc.)
If I understand correctly, the number inDense(3...) defines the amount of classes.

outputs1=Dense(3,activation='softmax')(att_out) 
model1=Model(inputs1,outputs1)
model1.summary()
model1.compile(optimizer="adam", loss="sparse_categorical_crossentropy", metrics=['accuracy'])
model1.fit(x=text_pad,y=train_y,batch_size=batch_size,epochs=epochs,verbose=1,shuffle=True) 

y_pred = model1.predict(test_text_matrix)

Output example:

[[0.34014237 0.33570153 0.32415614]
 [0.34014237 0.33570153 0.32415614]
 [0.34014237 0.33570153 0.32415614]
 [0.34014237 0.33570153 0.32415614]
 [0.34014237 0.33570153 0.32415614]]

Output I want:

[1,2,0,0,0,1,2,0]

Thank you for any ideas.

Answers:

Thank you for visiting the Q&A section on Magenaut. Please note that all the answers may not help you solve the issue immediately. So please treat them as advisements. If you found the post helpful (or not), leave a comment & I’ll get back to you as soon as possible.

Method 1

You did not do anything wrong, predict has always returned the output of the model, for a classifier this has always been probabilities per class.

predict_classes is only available for Sequential models, not for Functional ones.

But there is an easy solution, you just need to take the argmax on the last dimension and you will get class indices:

y_probs = model1.predict(test_text_matrix)
y_pred  = np.argmax(y_probs, axis=-1)

All methods was sourced from stackoverflow.com or stackexchange.com, is licensed under cc by-sa 2.5, cc by-sa 3.0 and cc by-sa 4.0