Feature request

[saersamani]

Hi,

I have been using this package with very large datasets and I very much like it. When saving the models for the deployment the final file (i.e. the model) is extremely large (e.g. 6 GB) even with the compression provided by JLD2.

I was wondering if there is another way to save the models into smaller files or only saving the model without the data itself.

Thank you in advance.

Cheers,
Saer

[bensadeghi]

Not much can he done here, without major changes to the code.

When building models, make sure to cast the training data to explicit types (ie with float.(), string.(), etc). This significantly improves model training and prediction execution times, and also drastically reduces the size of saved models.