Improve distributed tests for Confusion Matrix (CM and MLCM)

[vfdev-5]

Currently, the testing data is generated without taking into account distributed context:

ignite/tests/ignite/metrics/test_confusion_matrix.py

Lines 559 to 565 in e17acc7

	def _test_distrib_multiclass_images(device):
	def _test(metric_device):
	num_classes = 3
	cm = ConfusionMatrix(num_classes=num_classes, device=metric_device)
	y_true, y_pred = get_y_true_y_pred()

same for MultiLabelConfusionMatrix tests.

We would like to rewrite the tests in a similar way as here :

ignite/tests/ignite/metrics/test_recall.py

Lines 778 to 785 in e17acc7

	def _test(average, n_epochs, metric_device):
	n_iters = 60
	s = 16
	n_classes = 7
	offset = n_iters * s
	y_true = torch.randint(0, 2, size=(offset * idist.get_world_size(), n_classes, 6, 8)).to(device)
	y_preds = torch.randint(0, 2, size=(offset * idist.get_world_size(), n_classes, 6, 8)).to(device)

See https://github.com/pytorch/ignite/issues#issuecomment-782750122

cc @touqir14

[Moh-Yakoub]

I saw this issue and couple of PRs linked to it. Is this fully done already?

[vfdev-5]

No it is still to be done

[touqir14]

I can start working on this but I am a bit busy at the moment and not before 4-5 days unfortunately. If @Moh-Yakoub wants to take a shot, that is fine.

[Moh-Yakoub]

I think I can start working on this.

[Moh-Yakoub]

@vfdev-5 I feel I am missing something here about the tests. Based on investigating the test cases in the confusion matrix file. I found that the distributed context was already based in the distributed test cases

For example:

def test_distrib_gpu(local_rank, distributed_context_single_node_nccl):

    device = torch.device(f"cuda:{local_rank}")
    _test_distrib_multiclass_images(device)
    _test_distrib_accumulator_device(device)

[vfdev-5]

@Moh-Yakoub yes, there are distributed tests, it is what I say in the issue description. The problem is that the data is generated without taking into account distributed context. Currently, the data is the same for all procs. What we want is to generate data dependant of the proc rank. Please, reread issue description for details.