Add regression benchmark about pmlb HOT 15 CLOSED

Ok will do:-)

from pmlb.

I tried to convert those regression data to tsv file in pmlb But I did not merge them into pmlb repo yet. I found a few dataset is very huge, like >200Mb in tsv format. You can check them in the LPC path I sent over in slack. Should we exclude them?

Another issue with pmlb. The target name of all current classification benchmarks in pmlb is class, I think it is not good for regression benchmark so I used pmlb_reg_target instead, is it OK?

from pmlb.

How big are the files when gzipped?

Let's call the class equivalent as target.

from pmlb.

Even gzip file is more than 110Mb. It is a benchmark with 1 million samples.

target sounds good

from pmlb.

We might end up changing the PMLB standard such that all target columns are named target.

from pmlb.

if you want i can add a collection of benchmark problems used in GP (mostly synthetic).

from pmlb.

@lacava sounds good.

from pmlb.

Sure @lacava, send over a PR and describe them for us. 👍

from pmlb.

i was thinking it would also be nice to have a notebook for generating the synthetic datasets, for example for the korns / keijzer / pagie problems. where should that go?

from pmlb.

Maybe that would be a separate project, unless we were explicitly creating datasets to upload to PMLB.

from pmlb.

yea, they would correspond to datasets in the archive.

from pmlb.

Cool, send a separate PR for it and we'll take a look. :-)

from pmlb.

@weixuanfu, can you please re-send the PR with the regression benchmarks? Please take care when moving the existing classification datasets around.

from pmlb.

OK, will do

from pmlb.

Please check PR #15

from pmlb.