Errror using custom loss

[max37400]

Hello!

I've tried to use custom loss function like this:

def rmlse(y_true, y_pred):
    return mean_squared_logarithmic_error(y_true, y_pred)**0.5

optimizer = keras.optimizers.Adam(learning_rate=0.05)
model.compile(optimizer=optimizer, loss=rmlse, metrics=['mean_absolute_error'])

The model was built successfully, but when we tried to upload that model with mlem, the provided error rises:

Unexpected error: Unknown loss function: 'rmlse'. Please ensure you are using a keras.utils.custom_object_scope and that this object is included in the scope. See 
https://www.tensorflow.org/guide/keras/save_and_serialize#registering_the_custom_object for details.

How can I use the custom loss functions? Thanks!

[aguschin]

I assume this happens when trying to load the model. Looks like MLEM does not pin down this requirement when recursively going over the model at saving, I guess we need to see if this is fixable as a special case + add some option to make MLEM investigate reqs deeper (get more broad list of reqs, but at the same time pin reqs like this down).