model.fit works in colab but not in PyCharm

[yucesoye]

CNN_SpeechEmotion.ipynb works in colab but I can't run it with python (with pyCharm) that I installed locally.
Why is this?

The result I got on Colab:
11/11 [==============================] - 0s 6ms / step - loss: 0.4264 - accuracy: 0.8207
Restored model, accuracy: 82.07%

I have a few more questions about the code
I will be very glad if you help

1. Although 4 classes were used in the project ("angry", "sad", "neutral", "happy"), 8 neurons were used in the dense layer.

Shouldn't dense layer have the same number of neurons as the number of classes?

2. Some elements of x_traincnn used in training the model are 0 (especially after 150).
   In a question asked before, it was said that instead of 0s, it was necessary to use averages.

Is it relative to the average row or column? (time / attribute)
3)
Can you share the X_train, X_test, y_train, y_test data that you have prepared for training and testing the model? I will compare it to my data

np.save ('X_train', X_train)
np.save ('X_test', X_test)
np.save ('y_train', y_train)
np.save ('y_test', y_test)