Use serialization library for storing / loading state of pipeline modules?

[damellis]

It might be nice to use something like Cereal or boost::serialization for saving and loading the state of pre-processing, feature extraction, classification modules, etc. That would simplify the process of writing new modules by reducing the amount of boilerplate saving and loading code required.

[nickgillian]

Yes, this would be great, particularly for reducing the boilerplate code required for writing new modules.

The key thing here is how to add this sort of support without bringing in a major dependency like boost (a great library, but I'm trying to keep the toolkit free from major dependencies like this).

[damellis]

Good point. Cereal seems relatively light-weight: it looks like its headers-only and doesn't require any dependencies. It's still an external code-base, but it's a relatively small one.

[cyberluke]

I'm now doing this manually - simple txt file with parameters to save it together with training data.