I have added a global for the default evaluator to which @vfdev-5 was referring to in this comment.
I also felt we could mention it somewhere... What could be a right place to mention it, if necessary?
Thanks :)

@DevPranjal Thank you. I think we already discussed that point, and the idea is to have simple and copy/paste pieces of code, even if a lot of code could be reused.

@vfdev-5 what do you think ?

@sdesrozis I do not see the point to make API examples cluttered with other "unrealistic" code just to make them executable.
We can however publicly display doctest_global_setup to users such that they could reuse it if needed reproducible code

The point is the user can't copy/paste the code to test. I thought we would like to provide users with easily reproducible examples.

Anyway, let's choose whether we hide things and let's deploy 😊

I know, but it means for this metric we'll have the following:

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)

def process_function(engine, batch):
    y_pred, y = batch
    return y_pred, y

default_evaluator = Engine(process_function)

# --- API example 
metric = MeanPairwiseDistance(p=4)
metric.attach(default_evaluator, 'mpd')
preds = torch.Tensor([
                [1, 2, 4, 1],
                [2, 3, 1, 5],
                [1, 3, 5, 1],
                [1, 5, 1 ,11]
])
target = preds * 0.75
state = default_evaluator.run([[preds, target]])
print(state.metrics['mpd'])

You have half of the code snippet of preps.

We are talking about more or less 30 code lines, and using lambda function, we could reduce this. Not sure that it's worth hiding rather than exposing.

Anyway, that's a matter of taste.