Alex Meiburg

Err, a correction: it should be masked with valid moves, *and then re-normalized*. Otherwise it's not changing anything.

For reference: I fixed this bug in my local copy, and it appears to be learning _several times faster_ . Passing in the valid moves requires adapting a lot of...

I realized that my wording above was somewhat vague. `pi = valids * pi` works, but then you can no longer use `log_softmax`, since that has the normalization inside of...

I had this issue as well -- appeared to happen whenever I mixed Core.Tuple and Core.Array in the type, e.g. `Core.Array{Core.Tuple{Core.Int64, Core.Array{Core.Float64}}}`. Updating JLD `v0.10.0 ⇒ v0.12.3` in Pkg --...

It seems the problem happens when you build SCIP once without setting the SCIPOPTDIR environment variable; subsequently rebuilding doesn't automatically fix it. Doing `Pkg.rm("SCIP")` isn't enough. Here's what I did...

I seem to be having a similar issue, with presolving leading this message and then HiGHS complaining about an infeasible problem. Unlike issue #771, I don't think has anything to...

Sure, here: [bad_highs_model.txt](https://github.com/ERGO-Code/HiGHS/files/11265729/bad_highs_model.txt) That's the .lp format, but GitHub insisted I give it a known file extension (.txt). For reference, I'm getting this error on HiGHS v1.5.1, githash 93f1876e4, which...

Yes, on Windows. I could try it without the Julia wrapper, later, to see if that's relevant. (I would be suspicious of double values changing slightly or something, but given...

EDIT: I think this is an easier way to replicate the issue locally. Download [highs_file.txt](https://github.com/ERGO-Code/HiGHS/files/11267626/highs_file.txt), then run this Julia code: ``` using JuMP import HiGHS new_model = read_from_file("highs_file.txt"; format =...