Customized function takes forever to run for SymbolicTransformer about gplearn HOT 9 CLOSED

No, that is not normal. Can you give me more details?

• What does "take forever" mean?
• What version of gplearn are you running?
• Are you running all the commands in the notebook sequentially without modification?

from gplearn.

Just released the new version, can you try again with the latest? Let me know more about your error/environment/etc?

from gplearn.

Trevor, thanks a lot for your response. Somehow I didn't receive any notifications from Git for your replies. I will download the latest version and give it a try. I am running Python3.4 on Windows 7 machine with 8 cores. Hope this helps.

from gplearn.

Trevor, I tried the latest version and still have the same issue. Right now I use the default function set for now. I do have an extra question: I know gp._programs stores the symbolic expressions for each generation. Assume that I set population_size = 800, generations = 40, n_components = 20, and want to output the expressions of final results, shall I output gp._programs[39][0:19]?

Thank you so much for your patience.

from gplearn.

Update: we tested the same code on Linux environment and the same user-defined function takes No time to run. The same code ran for days on Windows without generating even the first iteration. There's no error information either. It's indeed related to the environment.

from gplearn.

Thanks for the follow up @hubokitty ... I'll attempt to set up Appveyor to run tests on Windows to see if it can replicate the problem.

Are you using a distribution like anaconda to install python and the numpy stack?

from gplearn.

SymbolicTransformer stores the final solutions in the attribute _best_programs. _programs contains all of the generations but it does not sort them so the [0:19] in your code will just get a random selection of the final generation.

from gplearn.

@hubokitty ... Tests are passing on Windows. I added a new test in #70 that performs the example in the docs with the logical function. It completes just fine on AppVeyor which runs on Anaconda in Windows.

The only difference I see that might be an issue for you is that that example uses the n_jobs attribute which invokes the joblib library. Can you try to run some other tests using joblib and see if that is the issue. If that is the case then the problem lies upstream of this package.

from gplearn.

Trevor, thanks a lot for your reply. I saw you opened another issue related to this. I look forward to the solution.

from gplearn.