`breedthreads = auto` causes individuals getting lost, and few exceptions

[ZvikaZ]

Hi.

When I add breedthreads = auto, I see that some individuals are getting lost, and there are few exceptions. (evalthreads = auto is working well).

I've encountered this issue before, in a larger project; but it also appears in the small sample project that I've just created: https://github.com/ZvikaZ/ECJ-sample (see https://github.com/GMUEClab/ecj/issues/76). Just run it, and it shortly fails. You can comment the breedthreads = auto in the params file, and see that everything is fine.

BTW, I have encountered this on two different machines, running Linux and Windows.

[eclab]

There are two bugs. What you see is a race condition. Fixing that revealed an underlying architectural issue with GroupedProblemForm which I have to think about, but in the meantime I've patched it with a hack.

Grab the following files from the repository and you should be good to go.

M src/main/java/ec/breed/BufferedBreedingPipeline.java M src/main/java/ec/simple/SimpleEvaluator.java

Sean

On May 23, 2021, at 8:20 AM, Zvika @.***> wrote:

Hi.

When I add breedthreads = auto, I see that some individuals are getting lost, and there are few exceptions. (evalthreads = auto is working well).

I've encountered this issue before, in a larger project; but it also appears in the small sample project that I've just created: https://github.com/ZvikaZ/ECJ-sample (see #76). Just run it, and it shortly fails. You can comment the breedthreads = auto in the params file, and see that everything is fine.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[ZvikaZ]

Indeed, it seems to solve the problem. Thanks for the quick fix. I got a lot of Yo mama prints, but as far as I understand the code that issues them, it should be harmless. Right?

BTW, it'd be nice to add this scenario to the tests, to avoid similar bug in the future.

[SigmaX]

BTW, it'd be nice to add this scenario to the tests, to avoid similar bug in the future.

Agreed! Your sample project should give us enough detail to create a test. Is it as simple as setting breedthreads = auto and watching the population size for a few generations?

Thanks, Siggy

On Mon, May 24, 2021 at 4:31 AM Zvika @.***> wrote:

Indeed, it seems to solve the problem. Thanks for the quick fix. I got a lot of Yo mama prints, but as far as I understand the code that issues them, it should be harmless. Right?

BTW, it'd be nice to add this scenario to the tests, to avoid similar bug in the future.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/GMUEClab/ecj/issues/77#issuecomment-846877568, or unsubscribe https://github.com/notifications/unsubscribe-auth/AACMOQQJW6E3LXPJCJMH4HLTPIFENANCNFSM45LVA5EA .

--

——

Eric "Siggy" Scott | MITRE

Senior Artificial Intelligence Engineer, MITRE Labs AI & Autonomous Systems Dept.

Doctoral Candidate, George Mason University http://mason.gmu.edu/~escott8/

[ZvikaZ]

BTW, it'd be nice to add this scenario to the tests, to avoid similar bug in the future.

Agreed! Your sample project should give us enough detail to create a test. Is it as simple as setting breedthreads = auto and watching the population size for a few generations? Thanks, Siggy

Yeah. Note that it already contains a checker for this: https://github.com/ZvikaZ/ECJ-sample/blob/124008a98dcfca0c6d3378364f261879409a7409/src/main/java/SampleProblem.java#L22

    var p = state.population.subpops.get(0);
    if (p.initialSize != p.individuals.size()) {
        state.output.fatal("someone got lost!!! (you might want to comment `breedthreads = auto` in the params file)");
    }

Also note, that now my sample project doesn't fail, as it's using a .jar that I've compiled with the fixes suggested in this thread.

[ZvikaZ]

There are two bugs. What you see is a race condition. Fixing that revealed an underlying architectural issue with GroupedProblemForm which I have to think about, but in the meantime I've patched it with a hack.

Grab the following files from the repository and you should be good to go.

M src/main/java/ec/breed/BufferedBreedingPipeline.java M src/main/java/ec/simple/SimpleEvaluator.java

Sean

Hi. Any news with the long-term solution? I think that I've found an issue with the hack, but I'm not sure it's worth reporting, if that hack will be replaced soon...

Thanks

[eclab]

This summer has presented time challenges to fixing bugs ECJ; I'm sorry for the extremely slow turnaround. I hope to get to it soon but may need more reminders. :-) In the meantime, go ahead and report issues with the hack.

Sean

On Aug 18, 2021, at 1:25 PM, Zvika @.***> wrote:

There are two bugs. What you see is a race condition. Fixing that revealed an underlying architectural issue with GroupedProblemForm which I have to think about, but in the meantime I've patched it with a hack.

Grab the following files from the repository and you should be good to go.

M src/main/java/ec/breed/BufferedBreedingPipeline.java M src/main/java/ec/simple/SimpleEvaluator.java

Sean

Hi. Any news with the long-term solution? I think that I've found an issue with the hack, but I'm not sure it's worth reporting, if that hack will be replaced soon...

Thanks

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://secure-web.cisco.com/1PXo6vsOT7PDAe1adY1vPZNo9tb3RzvaB2o2sk-8IGsxhCyBSlVwZSmsSgwObv9Cm30qC1BxO_Wakd4b2k-WbhRc8_AlBigzHcJeaJL-5QvSuFaqyuzgo3coDp1swaQuWXacSZ8H0BgTevHujCA7ToObX8POrjpbG-DM-4SwP9One3M9fU1P-KLMp6cdQXQaFz_SymL089gEniYnfZLvkLcfu4jfuXcVV52K3sTVePxX_FMOLQk4OTOJ07nbuAxfrZLmVSxBG-_3Tys_Sn47RnWKW8dMBOIwYfVL1rTH84y0dv9wqmW4a3jcuaUAULjaRQpmYiPYRKc37bHrvZRWb7u7X51IwvE9Vx-NlvyXAya-848FrU5ci6e1Xv2-XVg9GcR5xG3T8LMnN72KYdp1wE-qCYhsxjzEittQcTQWWSbPX0G9VmtEHTvURUvQYUU7r/https%3A%2F%2Fgithub.com%2FGMUEClab%2Fecj%2Fissues%2F77%23issuecomment-901294415, or unsubscribe https://secure-web.cisco.com/1DsFBwffdgdzrSU3DJ-Fj7jJfY3o3A_kGYX5jdnRWN5XC0P8ot34EjUlG46jj8bJzlnfp9bsSctuGOOYD863dZLmLJxFBzAAOpsnF4IcsqPQnEbL5YN7o1htIcRYd1ie3vMtjcO6q7w-T5aK1M3nJqPmpHokyNTOIs1fgRIsd9hyedOgpF1u1dWwvLs5B2Fv!%20_hBIFuvLteeRa-V1KstS4K9Y59Q3H0161susvnE36xPqmvylwoIOdyMiphHr1dnBJVxOs3b4p5fGg7i42umGdxV9QaqYA5BaOBDNVxIxiXoO3Zbsylh7LkP7GjcJkmLIZBOnylyky6PRNOD201zJCT3l6HSi1B0Q_XqcFMfU_wIfqWCXcCUT60eMGlo62SrBOs3hNZgPdulfb8gZ2Zva0XHrSVIpDeYvVZSBtl-hNlWLL735KrU00ki2ZfzTBsci-/https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FADAZDVEISC2J57MKVTESQW3T5PUKPANCNFSM45LVA5EA. Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email.

[ZvikaZ]

Great. I didn't get an answer for my 2 PRs since May (and I have reminded in https://github.com/GMUEClab/ecj/pull/79 few days ago), so I was afraid that the project has gone sleeping :-)

We're all busy, it's OK. I didn't want to be a nudge, so I hesitated with this.

It's a little bit related, since I'm working with a local version with those 2 PRs integrated. Thus, if you think that they will be reviewed soon, I prefer to report the details of the issue after that, so we will have the same version. Otherwise, I will run with the official master, to eliminate such influences.

[eclab]

I've decided that the hack is in fact probably good enough for long-term. I modified it slightly (hope I didn't break anything) and have committed it with a bit of documentation as well. Reopen the issue if things aren't working right.