Add support for explicit data
Closed this issue · 0 comments
nathancooperjones commented
Is your feature request related to a problem? Please describe.
Currently, Collie only supports implicit data (a user interacts with an item without assigning it an explicit rating), which works for most real-world recommendation problems. However, when we have a dataset with explicit ratings we'd like to use in Collie, we are left with a subpar model using less data than it could!
This is bad!
Describe the solution you'd like
Add in explicit data support! This would likely be through four separate pieces to match the work done with implicit data support:
- Add
ExplicitInteractions
andExplicitHDF5Interactions
datasets, returning ratings instead of negative items - Add explicit loss functions (NOTE: these do not need partial credit functionalities)
- Add explicit loss function calculation support in model classes
- Add explicit metrics calculated on the GPU
Describe alternatives you've considered
There is no alternative that satisfies me!
Any additional information?
Nope!