Added * reproducible results on a single machine via the `seeds` parameter. * proper CUDA support. Automatic detection if you have a CUDA-enabled machine.
0.2.6
Added * enable logistic output/binary classification. Occurs automatically when `torch.nn.BCELoss()` function passed to the `criterion` parameter.
0.2.5
Fixed * fixed `init_test_size` of model selection functions working on the full dataset including NAs in the target variable
0.2.4
Added
* make model instantiation and model selection robust to variables with no data in them
0.2.3
Added
* feature contribution weighting for data availabilty in interval predict functions
Fixed
* got rid of hardcoded references to _date_ column in interval predict code
0.2.2
Added
* ability to generate uncertainty intervals via the `model.interval_predict()` function * ability to generate uncertainty intervals on synthetic vintages via the `model.ragged_interval_predict()` function