Skip workflow run if no data provided

[Nic-Ma]

Is your feature request related to a problem? Please describe.
Feature request from @SachidanandAlle :
Earlier we were not able to support zero input for validation, when there is no items for validation, we like to skip the evaluator.
can we add that fix if it's not there? which is needed to run train/finetune over few images, because some-times user submits only few samples for (re)training, and won't provide samples for evaluation/validation step.

[Nic-Ma]

Program to reproduce the exception:

dataloader = []
evaluator = SupervisedEvaluator(
    val_data_loader=dataloader,
    device=torch.device("cpu"),
    epoch_length=0,
    network=TestNet(),
    non_blocking=False,
    prepare_batch=PrepareBatchDefault(),
    decollate=False,
)
evaluator.run()

output:

  evaluator.run()
File "/workspace/data/medical/MONAI/monai/engines/evaluator.py", line 137, in run
  super().run()
File "/workspace/data/medical/MONAI/monai/engines/workflow.py", line 275, in run
  super().run(data=self.data_loader, max_epochs=self.state.max_epochs)
File "/opt/conda/lib/python3.8/site-packages/ignite/engine/engine.py", line 684, in run
  raise ValueError("Input data has zero size. Please provide non-empty data")
ValueError: Input data has zero size. Please provide non-empty data

Thanks.

[SachidanandAlle]

This this is the issue.. and same in case of multi-gpu when you are splitting the data by N (number of gpus).. example. there are 3 validations samples.. and 4 GPUS.. first 3 gets 1 each.. and the last one have none..