fastFM
fastFM copied to clipboard
Compatibility with scikit-learn object API
The estimators do not seem to have full compatibility with scikit-learn object API, which is described here.
For example, fastFM.sgd.FMRegression
does not store l2_reg
in __init__
. This prevents sklearn.base.clone
to clone an estimator, as shown in the image below. Due to this problem, currently we cannot use some of sklearn's useful utilities such as GridSearchCV
.
As recommended in sklearn's coding guide, we might want to test the compatibility by using check_estimator
, and modify the estimators to pass the test.
Full API compatibility is certainly something we are aiming for. Could you list what else is missing?
Issue #49 is also related to incomplete API compatibility.