search

ODINN-SciML / MassBalanceMachine

Global machine learning glacier mass balance model, capable of assimilating all sources of glaciological and remote sensing data
MIT License
15 stars 11 forks source link

Create a wrapper class that includes data(?), model, and gridsearch #40

Open JulianBiesheuvel opened 3 months ago

JulianBiesheuvel commented 3 months ago

With a wrapper class, the user can choose what model they would like to use (for now it only accommodates XGBoost but in the future possibly a neural network), accounts for the grid-search (often the model and the grid-search go together), and the data (t.b.d.). It also provides extra functionalities like saving and loading a model from a file, and more...

JulianBiesheuvel commented 2 months ago

The architecture we discussed—having a parent class Model with child classes XGBoostCustom and MLP—is too simple and not feasible. We should consider a more elaborate approach, as I described earlier. For now, I've implemented the custom XGBoost class with all the necessary features. However, features like saving and loading the model and performing grid search should ideally be part of a wrapper class rather than being included in each individual model. Some features that are currently implemented for the CustomXGBoostRegressor (but should be part of a wrapper), are:

For now, this, and the issues listed above, are on hold. In future releases, a wrapper/parent class will be provided that accounts for different kinds of models. For now, the focus is on the XGBoost model.