tqdm_logger does not take epoch_length into account

[remisphere]

🐛 tqdm_logger does not take epoch_length into account

When calling Engine.run() with a custom epoch_length,
the tqdm progress bar does not adapt and displays the full number of batches in the data.
Here is a minimal example:

from ignite.contrib.handlers.tqdm_logger import ProgressBar
from ignite.engine import Engine
from torch import nn
from torch.utils.data import DataLoader

data = list(range(100))
model = nn.Identity()
engine = Engine(lambda engine, batch: model(batch))

ProgressBar(persist=True).attach(engine)
engine.run(data, epoch_length=50)

We have 100 items in data but the true end of the epoch is at 50 iterations, yet the progress is displayed over the range of 100 and just ends midway, when I expect it to be displayed over the range of 50, thus ending when the bar is full.
One can not overwrite tqdm's total argument by replacing

ProgressBar(persist=True).attach(engine)

by

ProgressBar(persist=True, total=50).attach(engine)

for it raises TypeError: type object got multiple values for keyword argument 'total'.

Environment

• PyTorch Version : 1.4.0
• Ignite Version : 0.3.0
• OS : Ubuntu 19.04
• Ignite installation method : pip
• Python version: 3.7.3

[vfdev-5]

@remisphere thanks for the feedback! We need to fix it here:

ignite/ignite/contrib/handlers/tqdm_logger.py

Lines 220 to 221 in 0f97baf

	if event_name in (Events.ITERATION_STARTED, Events.ITERATION_COMPLETED):
	return len(engine.state.dataloader)

as

if event_name in (Events.ITERATION_STARTED, Events.ITERATION_COMPLETED): 
    return engine.state.epoch_length