How to add custom metric?

[ArMaxik]

Hello!

I've implemeted my own Dataset class with overloaded evaluate() method, wich can calculate my custom metric. Then I've changed config:

dataset_type = 'MyDataset'
...
evaluation = dict(interval=1, metric=['f1_score', 'custom_metric', 'accuracy'], save_best='accuracy_top-1')

But during training custom metric has been calculated once, only after 1 epoch. Then it dissapeared and there are only 'f1_score', 'accuracy' in the log.

What i should do to have my new metric works?

[Ezra-Yu]

please use the 1.x branch. In the 1.x branch, this issue has been resolved.
Moreover, we are not going to spend a lot of effort on the master branch at the moment and will probably just fix some bugs later on, with all new features being developed on 1.x.