task_labels experience attribute type fix: now it is a list, not a set

[vlomonaco]

Hello, I found a bug in the task_incremental_benchmark generator. Even with the MNIST example in the notebook "03_benchmarks.ipynb" exp.task_labels is a set not a list as the train method expects.

I think the problem is here, where we convert the set in the list only for the task_label attribute:

avalanche/avalanche/benchmarks/scenarios/task_aware.py

Lines 89 to 105 in 0d6f715

	def with_task_labels(obj):
	"""Add `TaskAware` attributes.
	The dataset must already have task labels.
	`obj` must be a scenario, stream, or experience.
	"""
	def _add_task_labels(exp):
	tls = exp.dataset.targets_task_labels.uniques
	if len(tls) == 1:
	# tls is a set. we need to convert to list to call __getitem__
	exp.task_label = list(tls)[0]
	exp.task_labels = tls
	return exp
	return _decorate_generic(obj, _add_task_labels)

The set is never converted into a list for the attribute task_labels and in in this method we use the indexing which is not supported by a set:

avalanche/avalanche/evaluation/metric_utils.py

Lines 249 to 275 in 0d6f715

	def phase_and_task(strategy: "SupervisedTemplate") -> Tuple[str, int]:
	"""
	Returns the current phase name and the associated task label.
	The current task label depends on the phase. During the training
	phase, the task label is the one defined in the "train_task_label"
	field. On the contrary, during the eval phase the task label is the one
	defined in the "eval_task_label" field.
	:param strategy: The strategy instance to get the task label from.
	:return: The current phase name as either "Train" or "Task" and the
	associated task label.
	"""
	task_labels = getattr(strategy.experience, "task_labels", None)
	if task_labels is not None:
	task = task_labels
	if len(task) > 1:
	task = None # task labels per patterns
	else:
	task = task[0]
	else:
	task = None
	if strategy.is_eval:
	return EVAL, task
	else:
	return TRAIN, task

Below you can find the small fix + an assert I added in the TaskAware tests.

[coveralls]

Pull Request Test Coverage Report for Build 9193158769

Details

• 2 of 3 (66.67%) changed or added relevant lines in 2 files are covered.
• 1 unchanged line in 1 file lost coverage.
• Overall coverage remained the same at 51.803%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
tests/benchmarks/scenarios/test_task_aware.py	0	1	0.0%

Files with Coverage Reduction	New Missed Lines	%
tests/benchmarks/scenarios/test_task_aware.py	1	34.21%

Totals
Change from base Build 9014120340:	0.0%
Covered Lines:	15082
Relevant Lines:	29114

---

💛 - Coveralls