[skip ci] Doctest for Accuracy

[Chandan-h-509]

Addresses #2265

Description: Added doctests for 'Accuracy'

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)

[vfdev-5]

Thanks @Chandan-h-509 , I left a comment.

[vfdev-5]

I think we could create a default evaluator as above and hide this part of code from the user ?

cc @ydcjeff

[Chandan-h-509]

I feel it would be good to have this part of the code so that it remains transparent.
Could you also explain the alternative that you are proposing?

[vfdev-5]

There is a way to define a global context using this:

ignite/docs/source/conf.py

Lines 344 to 353 in 74dabca

	doctest_global_setup = """
	import torch
	from torch import nn, optim
	from ignite.engine import *
	from ignite.handlers import *
	from ignite.metrics import *
	from ignite.utils import *
	manual_seed(666)

[Chandan-h-509]

@vfdev-5 I have made the necessary change.
Kindly review

[vfdev-5]

@Chandan-h-509 thanks for the update, but (here I agree with your previous comment) it is non-obvious to understand what it is evaluator with its particular API as evaluator(Accuracy(), metric_name). I think a tradeoff would be something like

            metric = Accuracy()
            metric.attach(evaluator, "accuracy")

            preds = torch.tensor([[1,0,0,1], ])
            target = torch.tensor([[1,0,0,0], ])
            state = evaluator.run([[preds, target], ])
            print(state.metrics["accuracy"])

[vfdev-5]

In addition, I think it would be more helpful to show all possible options while computing accuracy:

1. binary case
2. multiclass case
3. multilabel case

[Chandan-h-509]

@Chandan-h-509 thanks for the update, but (here I agree with your previous comment) it is non-obvious to understand what it is evaluator with its particular API as evaluator(Accuracy(), metric_name). I think a tradeoff would be something like

            metric = Accuracy()
            metric.attach(evaluator, "accuracy")

            preds = torch.tensor([[1,0,0,1], ])
            target = torch.tensor([[1,0,0,0], ])
            state = evaluator.run([[preds, target], ])
            print(state.metrics["accuracy"])

@vfdev-5 I didn't understand the solution clearly.
Could you please elaborate more on it.

[vfdev-5]

@Chandan-h-509 current docstring you proposing in this PR does not look good especially this line:

engine=evaluator(Accuracy(), metric_name)

because it is not obvious to understand what it is evaluator method and what is its API.
I was proposing you to rework the docstring and doctest_global_setup such that the new docstring ressembles to https://github.com/pytorch/ignite/pull/2287/files#r735168912

Additionally, you need to add more docstrings for other cases as described https://github.com/pytorch/ignite/pull/2287/files#r735169293

[sdesrozis]

@Chandan-h-509 Thank you for this PR. Are you still working on it ?

[Chandan-h-509]

@sdesrozis sorry for the delay.. I have been very busy these last 20days with my college exams and other activities.. I will try to complete it in a few days time.

[sdesrozis]

@Chandan-h-509 Thank you for your help. I close this PR since a doctest for Accuracy has been merged.