Error when applying SoftTreeSupLoss to custom model/dataset about neural-backed-decision-trees HOT 5 CLOSED

Code looks good. What's the error? You can just run the model as an NBDT. However, fine-tuning with the SoftTreeSupLoss will improve accuracy drastically.

from neural-backed-decision-trees.

Hey
thanks for your reply
everytime i try to run the code for my custom dataset
i get an "assertion error"
below is the code:
from nbdt.loss import SoftTreeSupLoss
criterion = nn.CrossEntropyLoss()
criterion = SoftTreeSupLoss(dataset=data['train'], criterion=criterion)

where data['train'] is a dataloader for my dataset

Sorry for the confusion but please let me know if i am missing anything

Error snipet:
/usr/local/lib/python3.6/dist-packages/nbdt/data/custom.py in dataset_to_dummy_classes(dataset)
31
32 def dataset_to_dummy_classes(dataset):
---> 33 assert dataset in DATASETS
34 num_classes = DATASET_TO_NUM_CLASSES[dataset]
35 return [FakeSynset.create_from_offset(i).wnid for i in range(num_classes)]

from neural-backed-decision-trees.

I see, just modify the nbdt.utils.DATASETS constant to include your custom dataset. Also, dataset section of the README will be helpful: https://github.com/alvinwan/neural-backed-decision-trees#dataset

from neural-backed-decision-trees.

Hi,
thanks a lot for your reply and time.
I modified the utils.py and modified the datasets too for my custom dataset.
However, can you guide me to execute it?
The reason being, after modifying both the python files I get the same error.
After modifying both files, is their a different way to run these files?

from neural-backed-decision-trees.

Try steps 3 and 4 (I just added these) in the README "datasets" section. It now elucidates which constants in nbdt.utils to modify.

from neural-backed-decision-trees.