ni_benchmark with 2D target values

[Mpkosh]

Hello!

I'm having trouble creating ni_benchmark for data, where target values are 2D.
As far as I understand, avalanche "compresses targets and task labels to save some memory". Is there a way to solve this, use np.vectorize instead or disable this compression?

all_x.shape
(100717, 60, 18)

all_y.shape
(100717, 7, 14)

Creating ni_benchmark:

scenario = ni_benchmark(
    TensorDataset(torch.tensor(all_x, dtype=torch.float16), 
                  torch.tensor(all_y, dtype=torch.int8)), 
    TensorDataset(torch.tensor(all_test_x, dtype=torch.float16), 
                  torch.tensor(all_test_y, dtype=torch.int8)), 
    n_experiences=10, shuffle=True, seed=1234,
    balance_experiences=True
)

Error:

c:\users\user\anaconda3\lib\site-packages\avalanche\benchmarks\utils\dataset_utils.py(85)__getitem__()
     83 
     84         if self.converter is not None:
---> 85             return self.converter(element)
     86 
     87         return element

ipdb>  self.converter
<class 'int'>
ipdb>  element
tensor([[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0, 0, 0, 1],
        [0, 0, 0, 1, 1, 0, 1, 0, 0, 1, 1, 0, 0, 0],
        [0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1, 1],
        [0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        [0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 1, 0, 0, 0],
        [0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]], dtype=torch.int8)

Full error:

ValueError                                Traceback (most recent call last)
Input In [135], in <cell line: 1>()
----> 1 scenario = ni_benchmark(
      2     TensorDataset(torch.tensor(all_x, dtype=torch.float16), 
      3                   torch.tensor(all_y, dtype=torch.int8)), 
      4     TensorDataset(torch.tensor(all_test_x, dtype=torch.float16), 
      5                   torch.tensor(all_test_y, dtype=torch.int8)), 
      6     n_experiences=10, shuffle=True, seed=1234,
      7     balance_experiences=True
      8 )

File ~\anaconda3\lib\site-packages\avalanche\benchmarks\generators\benchmark_generators.py:347, in ni_benchmark(train_dataset, test_dataset, n_experiences, task_labels, shuffle, seed, balance_experiences, min_class_patterns_in_exp, fixed_exp_assignment, train_transform, eval_transform, reproducibility_data)
    342 transform_groups = dict(
    343     train=(train_transform, None), eval=(eval_transform, None)
    344 )
    346 # Datasets should be instances of AvalancheDataset
--> 347 seq_train_dataset = AvalancheDataset(
    348     seq_train_dataset,
    349     transform_groups=transform_groups,
    350     initial_transform_group="train",
    351     dataset_type=AvalancheDatasetType.CLASSIFICATION,
    352 )
    354 seq_test_dataset = AvalancheDataset(
    355     seq_test_dataset,
    356     transform_groups=transform_groups,
    357     initial_transform_group="eval",
    358     dataset_type=AvalancheDatasetType.CLASSIFICATION,
    359 )
    361 return NIScenario(
    362     seq_train_dataset,
    363     seq_test_dataset,
   (...)
    371     reproducibility_data=reproducibility_data,
    372 )

File ~\anaconda3\lib\site-packages\avalanche\benchmarks\utils\avalanche_dataset.py:296, in AvalancheDataset.__init__(self, dataset, transform, target_transform, transform_groups, initial_transform_group, task_labels, targets, dataset_type, collate_fn, targets_adapter)
    290 """
    291 The collate function to use when creating mini-batches from this
    292 dataset.
    293 """
    295 # Compress targets and task labels to save some memory
--> 296 self._optimize_targets()
    297 self._optimize_task_labels()
    298 self._optimize_task_dict()

File ~\anaconda3\lib\site-packages\avalanche\benchmarks\utils\avalanche_dataset.py:1115, in AvalancheDataset._optimize_targets(self)
   1114 def _optimize_targets(self):
-> 1115     self.targets = optimize_sequence(self.targets)

File ~\anaconda3\lib\site-packages\avalanche\benchmarks\utils\dataset_utils.py:424, in optimize_sequence(sequence)
    421 if isinstance(sequence, list):
    422     return sequence
--> 424 return list(sequence)

File ~\anaconda3\lib\_collections_abc.py:874, in Sequence.__iter__(self)
    872 try:
    873     while True:
--> 874         v = self[i]
    875         yield v
    876         i += 1

File ~\anaconda3\lib\site-packages\avalanche\benchmarks\utils\dataset_utils.py:85, in SubSequence.__getitem__(self, item_idx)
     82 element = self._targets[subset_idx]
     84 if self.converter is not None:
---> 85     return self.converter(element)
     87 return element

ValueError: only one element tensors can be converted to Python scalars