How to get the output of the model, before the softmax, without chaning the model architecture?

TensorFlow General Discussion
,
1

I have a trained sequential keras model.

The last layer is a Dense layer with softmax activation function:

model = keras.models.Sequential()
model.add(...)
model.add(...)
model.add(...)
model.add(keras.layers.Dense(50, activation='softmax'))

How can I get the the output of the model, before the softmax, without changing the model architecture ? (I have trained model, which I can’t change or train).

I have tried with:

probs = model.predict(X_train)
logits = probs - np.log(np.sum(np.exp(probs), axis=-1, keepdims=True))

but it seems that if I’m running softmax on logtis, it give me different results from probs:

def softmax(x):
    e_x = np.exp(x - np.max(x))  
    return e_x / e_x.sum(axis=1, keepdims=True)

probabilities = softmax(logits)
3

can you try this?
model_without_last_layer = Model(inputs=model.input, outputs=model.layers[-2].output)