data loading: fix a bug in choosing the data sampler

_get_sampler returned None if fixed_subset == False.
This was caused by a wrong indentation level such that
the case of fixed_subset == False was not handled leading
to _get_sampler returning None.

The bug was introduced yesterday, in commit
eb4da48b

distiller/apputils/data_loaders.py

@@ -265,7 +265,7 @@ def _get_sampler(data_source, effective_size, fixed_subset=False, sequential=Fal
             return SubsetSequentialSampler(subset_indices)
         else:
             return torch.utils.data.SubsetRandomSampler(subset_indices)
-        return SwitchingSubsetRandomSampler(data_source, effective_size)
+    return SwitchingSubsetRandomSampler(data_source, effective_size)
 
 
 def get_data_loaders(datasets_fn, data_dir, batch_size, num_workers, validation_split=0.1, deterministic=False,