Kernel appears to have died in Jupyter Notebook using TPOTClassifier

[ben0it8]

Hello,

When calling the fit method from a Jupyter Notebook, the kernel seems to die.
I use Python 2.7 with the latest Jupyter release and an upgraded version of tpot.
First I have this DeprecationWarning:

Then, after a couple of seconds I get the following window popping up:

Do you guys have any idea what could be the reason?

[weixuanfu]

I think it may be related to the issue #508. Please try to set n_jobs=1 with python 2.7 or try use forkserver start mode in python 3.4+ (see the demo codes in issue #508)

[jonathanng]

I'm getting the same thing with TPOTRegressor.

[weixuanfu]

jonathanng Could you please let me know whether setting n_jobs=1 or using forkserver start mode in python 3.4+ in TPOTRegressor can solve this issue? Below is a demo for TPOTClassifier.

import multiprocessing
if __name__ == '__main__':
    multiprocessing.set_start_method('forkserver')
    # Note: need move import sklearn into main unless a RuntimeError (RuntimeError: context has already been set) will raise
    from sklearn.datasets import make_classification
    from tpot import TPOTClassifier
    # make a huge dataset
    X, y = make_classification(n_samples=50000, n_features=200,
                                        n_informative=20, n_redundant=20,
                                        n_classes=5, random_state=42)
    
    # max_eval_time_mins=0.04 means 2 seconds limits for evaluating a single pipeline 
    # working in python 3.4+ in Linux OS and MacOS
    tpot = TPOTClassifier(generations=5, population_size=50, offspring_size=100, random_state=42, n_jobs=2, max_eval_time_mins=0.04, verbosity=3) 
    tpot.fit(X, y)

[jonathanng]

I'm using Python 3.6.1. I tried both n_jobs=1 and n_jobs=-1. Both gave me the error.

Also, when I run:

import multiprocessing
multiprocessing.set_start_method('forkserver')

I get the following error:

RuntimeError: context has already been set

I'm trying your timeout_pipe branch now. Will you let you know the results.

[weixuanfu]

I met this error before. You need restart the terminal and then start with:

import multiprocessing
if __name__ == '__main__':
    multiprocessing.set_start_method('forkserver')
    # Note: need move import sklearn into main unless a RuntimeError (RuntimeError: context has already been set) will raise

[weixuanfu]

@jonathanng please check the comment above.

[jonathanng]

This is all within a Jupyter notebook, so I don't think: if __name__ == '__main__': is necessary, right?

[weixuanfu]

@jonathanng oh you are right. It is not necessary in a jupyter notebook.

[jonathanng]

For both n_jobs=1 and n_jobs=-1, the timeout_pipe branch works.

Will this be merged into master?

[weixuanfu]

How about using forkserver start mode in current master branch?

The timeout_pipe branch is a compromising solution, and I tested it had some issues in Windows OS and also the performance of n_jobs was not as good as current master branch. We need refine this solution before merging if forkserver does not work or we need find a better solution to solve this fork issue.

[jonathanng]

With the timeout_pipe branch, it ran successfully, resulting in RandomForestRegressor as the best. But then I ran the data using XGBoost, and I get better results.

Did all the XGBoost runs fail in the pipeline_optimizer. How would I know if they didn't work.

[rhiever]

You can inspect all of the pipelines that were successfully evaluated during the run with the evaluated_individuals_ property of your TPOT object. Does XGBoost show up in any of the pipelines?

[jonathanng]

Ok, yes, I do see it. Thanks!

[jonathanng]

Everything ran successfully after setting n_jbos to half the # of processors and converting all the data to numpy arrays before feeding it into fit.

forkserver was not necessary.

[rhiever]

Interesting. Maybe it was a memory issue (not enough memory to run all jobs)?

[rhiever]

Closing this issue. Please feel free to re-open or file a new issue if you have any further questions or comments.

[vikeshsingh37]

I got same error when I installed xgboost. I tried all the solutions above but nothing worked for me. The problem disappeared after uninstalling xgboost. However, xgboost.XGBRegressor was not by TPOT.

[RamiroCope]

Thanks @vikeshsingh37 . This fixed the issue for me as well. Removing xgboost.XGBClassifier in my case did the job.