Postprocess output of model.predict()


I have an autoencoder that I have trained to encode and decode an image. After training I’m only using the encoder part of the autoencoder. The encoder outputs a feature vector. On inference (of the encoder) I need to compare the computed feature vector for some image with the mean feature vector computed from all training images. How can I adapt my model to do this step directly, so that the output of model.predict() returns the compared version of the feature vector and mean feature vector

I was looking into the possibility of overwriting the functionality of the model.predict() function. However, I don’t know if this is recommended.

What I’m doing now:

mean_feature_vector = (A np.array that consists of the mean feature vector for all n training images for the autoencoder)
feature_vector = model.predict(image)
# Compare feature vectors
comparison = np.linalg.norm(feature_vector-mean_feature_vector ) # The part that I want model.predict() to return

Hopefully my question makes sense, if not let me know :smiley: