Replace loss[0] with loss.item() due to deprecation

[sauercrowd]

When using tnt's meterlogger a deprecation warning by PyTorch does appear:

UserWarning: invalid index of a 0-dim tensor. This will be an error in PyTorch 0.
5. Use tensor.item() to convert a 0-dim tensor to a Python number

Issued by this line.

Replacing loss[0] with loss.item() should fix that without any further consequences

[nbro]

This also occurs with the MNIST with Visdom example. This is now an error in PyTorch 1.0.0.

[honghanchenchh]

Pytorch 1.1.0, error occurs with the mnist_with_meterlogger example
"IndexError: invalid index of a 0-dim tensor. Use tensor.item() to convert a 0-dim tensor to a Python number"
and it must fix This line

mnist_with_visdom example can be fixed
meter_loss.add(state['loss'].item())
# meter_loss.add(state['loss'].data[0])

[szagoruyko]

@honghanchenchh wasn't it fixed by #113 ?

[sauercrowd]

Yea it was, sorry forgot to close it.
Closing it for now

[banderlog]

It fixed in repo but not in python package torchnet-0.0.4 (last version)