search

mlr-org / ParamHelpers

Helpers for parameters in black-box optimization, tuning and machine learning.
https://paramhelpers.mlr-org.com
Other
25 stars 9 forks source link

Add option to configure the output length of the tuning info #204

Closed pat-s closed 5 years ago

pat-s commented 5 years ago

For example, Tune-x] 9: C=147; sigma=0.0615; fw.perc=0.463; fw.basal.methods=FSelector_re... should not be truncated.

Bernd said this is probably specified in https://github.com/berndbischl/ParamHelpers/blob/master/R/paramValueToString.R.

jakob-r commented 5 years ago

This would imply that we have to have some way to set options in ph. As this package is in maintenance mode from my point of view this feature will likely not be implemented. However, feel free to open a PR.

pat-s commented 5 years ago

Ok, no problem. Adding a cat() and using a fork does the job for me atm.

Will this option be available in paradox?

jakob-r commented 5 years ago

Logging is not done in ParamHelpers and as far as I can see also no truncation. Can you post this Issue in the package where it occurs, also including a minimal example or at least a more complete example of the output.

berndbischl commented 5 years ago

Jakob you are wrong the problem is in ph

berndbischl commented 5 years ago

I posted the explanation in slack. The problem is parValueToString. That's not really logging but being able to convert a PH list of values to a reasonable string. And that calls clipString, with a hard wired clip.len...

berndbischl commented 5 years ago

And such a function seems so fundamental that it SHOULD really exist in PH and probably also in paradox...

jakob-r commented 5 years ago

If you want to, open this issue again with a proposal what should be done.

berndbischl commented 5 years ago

well, just quickly: (and then maybe we can close again)

a) i completely agree that new development should be done in pd and not PH. so maybe the best thing is to do nothing here. also this is really not critical as in mlr you can write your own logger, and you have the OptPath with all the data.

b) Patrick is still right that this is bad what we did here (regarding a not so important thing..., see a)). and it is a problem that really is in PH.

c) for me the relevant question is what do in pd? dont we want to have a "toString" method there? then we should at least make some notes so the same mistakes doesnt happen again

thats it

berndbischl commented 5 years ago

so my suggestion is to simply discuss and agree here, then open an issue in pd?

jakob-r commented 5 years ago

I opened this issue: https://github.com/mlr-org/paradox/issues/43

berndbischl commented 5 years ago

I opened this issue: mlr-org/paradox#43

thx