ClassificationInterpreter AttributeError for custom dataset

I’m using fastai library with custom PyTorch dataset, which I specified the following way (... are there for the purpose of clarity):

from fastai import *

class CustomDataset(Dataset):
    def __init__(self, input_path):
    def __len__(self):

    def __getitem__(self, idx):
        return (
            torch.tensor(image, dtype=torch.float),
            torch.tensor(target, dtype=torch.long)

train_dset = CustomDataset(train_path)
valid_dset = CustomDataset(valid_path)

train_dl = DataLoader(train_dset, bs=16, shuffle=True)
valid_dl = DataLoader(valid_dset, bs=16, shuffle=True)

dls = DataLoaders(train_dl, valid_dl)

Then I’m using those dataloaders to train the model with Learner:

learn = Learner(dls, model, loss_func=CrossEntropyLossFlat(), metrics=accuracy)
learn.fit_one_cycle(1, 0.01)

and I pass learn object to ClassificationInterpretation class:

interp = ClassificationInterpretation.from_learner(learn)

However, this results in the AttributeError with not so clear details:

AttributeError                            Traceback (most recent call last)
<ipython-input-98-aa7f7b70a42b> in <module>
----> 1 interp = ClassificationInterpretation.from_learner(learn)

~/anaconda3/envs/python3/lib/python3.6/site-packages/fastai/ in from_learner(cls, learn, ds_idx, dl, act)
     40         _,_,losses = learn.get_preds(dl=dl, with_input=False, with_loss=True, with_decoded=False,
     41                                      with_preds=False, with_targs=False, act=act)
---> 42         return cls(learn, dl, losses, act)
     44     def top_losses(self, k=None, largest=True, items=False):

~/anaconda3/envs/python3/lib/python3.6/site-packages/fastai/ in __init__(self, learn, dl, losses, act)
     78     def __init__(self, learn, dl, losses, act=None):
     79         super().__init__(learn, dl, losses, act)
---> 80         self.vocab = self.dl.vocab
     81         if is_listy(self.vocab): self.vocab = self.vocab[-1]

~/anaconda3/envs/python3/lib/python3.6/site-packages/fastcore/ in __getattr__(self, k)
    387         if self._component_attr_filter(k):
    388             attr = getattr(self,self._default,None)
--> 389             if attr is not None: return getattr(attr,k)
    390         raise AttributeError(k)
    391     def __dir__(self): return custom_dir(self,self._dir())

~/anaconda3/envs/python3/lib/python3.6/site-packages/torch/utils/data/ in __getattr__(self, attribute_name)
     81             return function
     82         else:
---> 83             raise AttributeError
     85     @classmethod


It seems like both data loaders should have additional attribute vocab which might be there for DataBlock output, but not for my custom definition of data loaders. Is that really a problem? How this property should look like and how it could be added to make it compatible with ClassificationInterpreter?