Wrappers initialisation from full list of hyper-params - Githubissues

MobileTeleSystems / RecTools

RecTools - library to build Recommendation Systems easier and faster than ever before

Apache License 2.0

279 stars 36 forks source link

Wrappers initialisation from full list of hyper-params #157

Open blondered opened 3 months ago

blondered commented 3 months ago

Feature Description

Wrappers initialisation from full list of hyper-params (including wrapped)

Why this feature?

It will help us for save/load functions to models.
It will help running experiment from configs (not supported in framework but often used by users)
It will simplify code for experiments

Additional context

Interface is not decided. Options are:

model_factory method that accepts smth like:

model_type: ials
model_params:
fit_features_together: True
implicit_factors: 32
implicit_alpha: 10

accept params in __init__:
```
ials = ImplicitALSWrapperModel(fit_features_together=True, wrapped_params = {"factors: 32, alpha: 10"})
```
wrapped_params and initialised instance of wrapped model are both optional and user needs to pass one of them. If none is passed then default params are used.