Loaded Model Provides only Binary Outputs

General Discussion
#1

I trained a binary classification model that was providing predicted outputs within the range 0 - 1 (Sigmoid activation function outputting predictions). This was working as desired as below:

[[0.8134029 ]
 [0.20651695]
 ...
 [0.18252395]
 [0.87311405]]

However, after saving the model and trying to re-load and use it on the same data the output has changed to:

[[1.]
 [0.]
 ...
 [0.]
 [1.]]

Is there a way to return the output as described at the beginning when re-loading a saved model?

#3

@Jack_Zimmerman,

Welcome to the Tensorflow Forum!

I tried to save the model using model.save and reloaded using tf.keras.models.load_model and it is working as expected. How are you saving and reloading the model?

Could you please share the standalone code to reproduce the above issue?

Thank you!

#5

Answer was that i was just forgetting to normalise the input data based on my training data, just wanted to make sure I was making a mistake. Thanks for the help!

1 Like