What is the difference between ignite's metrics and pytorch's metrics?

[Jingnan-Jia]

I found that ignite have its own group of metrics. What is the difference with pure pytorch's metrics? Why does ignite not use pytorch's metrics directly?

update:
Here the pure pytorch's metrics mean pytorch's loss function.

[vfdev-5]

@Ordgod which "pure pytorch's metrics" you are talking about ?

[trsvchn]

@vfdev-5 I suppose loss functions? But we can use them with ignite's Loss metric. Am I right?

[vfdev-5]

Metrics were introduced to ignite almost since its begining. As far as I know, pytorch itself does not provide any metrics to compute. We did not introduced various SOTA loss functions as torch already has a number of built-in loss functions such that we wont overlap with it. Loss metric is a wrapper to compute average loss value for entire epoch (and not a single iteration).