Closed
Milestone
Oct 31, 2017–Nov 30, 2017
Restructure hiphive
The high level interface should be restructured to the discussed solution.
clusters = Clusters(prim, cutoffs)
sc = StructureContainer(clusters, supercells)
fj = Optimizer('LASSO',
sc.get_fit_data(property=..., structures=...),
validation_data=sc.get_fit_data(property=..., structures=...))
fj.train(training_fraction=0.5, validation_fraction=0.9)
model = clusters.get_model(fj.parameters)
This means we need the following core objects:
- Clusters (FCP minus parameters)
- StructureContainer (hiphive object minus parameters and train functionality)
- Optimizer (hiphive.train)
- Model (FCP)
And it would still be useful to have a FCM and FCS object.
Tasks:
-
Agree on this structure
✔ - Agree on suitable name for core classes
- Restructure code
- Update examples/tests
- Update doc/tutorial