ENH: Updated Loss metric to use required_output_keys

[01-vyom]

Fixes #1415

Description:
Similar to Metric, Loss metric now supports required_output_keys.

Check list:

• New tests are added (if a new feature is added)
• New doc strings: description and/or example code are in RST format
• Documentation is updated (if required)

[01-vyom]

Following is the demo code to check the validity of the above change:

import torch
import torch.nn as nn
from torch.nn.functional import nll_loss

from ignite.metrics import Accuracy, Loss
from ignite.engine import create_supervised_evaluator

model = nn.Linear(10, 3)

metrics = {
    "Accuracy": Accuracy(),
    "Loss": Loss(nll_loss)
}

# global criterion kwargs
criterion_kwargs = {"reduction": 'sum'}
# criterion_kwargs = {}

evaluator = create_supervised_evaluator(
    model,
    metrics=metrics,
    output_transform=lambda x, y, y_pred: {
        "x": x, "y": y, "y_pred": y_pred, "criterion_kwargs": criterion_kwargs}
)
data = [
    (torch.rand(4, 10), torch.randint(0, 3, size=(4, ))),
    (torch.rand(4, 10), torch.randint(0, 3, size=(4, ))),
    (torch.rand(4, 10), torch.randint(0, 3, size=(4, )))
]
res = evaluator.run(data)

As the required_output_keys contains criterion_kwargs, the user has to pass an empty dictionary for the original case with no criterion_kwargs for criterion.

[vfdev-5]

@01-vyom thanks for the draft PR !
Let's now add some tests here:

• https://github.com/pytorch/ignite/blob/master/tests/ignite/metrics/test_loss.py
  like here:

• ignite/tests/ignite/metrics/test_metric.py

  Line 98 in 692487c

  def test_output_as_mapping():

Instead of DummyMetric1, we can create a metric derived from Loss and override update method to check passed args similarly to here:

ignite/tests/ignite/metrics/test_metric.py

Lines 28 to 29 in 692487c

	def update(self, output):
	assert output == self.true_output

Please let me know if need more explanations.

We can also add your above demo example as an integration test to https://github.com/pytorch/ignite/blob/master/tests/ignite/metrics/test_loss.py

[01-vyom]

Ok, I will add a test_output_mapping as well as all other none_keys and wrong_keys test with a dummy Loss and an integration test similar to here:

ignite/tests/ignite/metrics/test_metric.py

Line 894 in 692487c

def test_override_required_output_keys():

[vfdev-5]

Thanks for the PR @01-vyom !