Daniel Ruffinelli

This is definitely a bug (whether in our configs or the code is a matter of interpretation). The idea was to have symmetric intervals, i.e. set a = -b, where...

I now think the cleanest thing is to just get rid of the automatic setting of non-given values and leave the rest as is. Sure, using lower_bound/upper_bound would be clearer...

Inconsistent how? Pytorch actually does not allow the lower bound to be higher than the upper bound (I had tested it wrong before), so we may just want Pytorch to...

There is no official implementation of triple classification yet, but the PRs you mention might have what you need. As for relation prediction, the reciprocal relations model does work in...

You can set the flag entity_ranking.metrics_per.head_and_tail to True and then all metrics are also reported by sp_ (tail) or _po (head). On Mon, 31 Oct 2022 at 06:00, Sirui ***@***.***>...

To me this is similar to the including test data during model selection: people do it but perhaps they shouldn't. While there is nothing wrong with selecting models based on...

In general, I also feel such things should be kept out the framework. Having used it hundreds of times on many different machines, this feels like something I, as a...