As this is a 2 year old issue and addresses v0.15, both GridSearchCV and joblib probably had significant changes since; @rasbt is this issue still relevant, or should it be closed?

Haven't tried this particular setting for a while. I would say let's close it due to the reasons you mentioned, but let me just try to re-run this in the next few days to double check.

I just ran the same code (the one that I posted on github in Dec 2014) over night and it seemed to be very stable this time (sklearn 0.18 and python 3.5) :); looks like the issue has been resolved some time ago!

thanks for checking :)

Hi @amueller , @rasbt ,
I've run into the same issue while trying to optimise hyper-parameters for a Random Forest Regressor.
I'm using Python 3.6 and sklearn version 0.20.3
My data set is not huge (~350k rows * 370 cols). Here's a snippet of the code that I'm using

    y_train = np.array(df_train[label])
    X_train = np.array(df_train[x_cols])
    weights = np.array(df_train['PREMISES_COUNT'])
    print('X_train shape ', X_train.shape)
    print('y_train shape ', y_train.shape)
    print('Sample weight shape ', weights.shape)
    
    param_grid = {'n_estimators': [100,150, 200],
                  'min_samples_leaf': [2,10,20],
                  'min_samples_split': [10,15,20],
                  'max_features': ['auto', 'sqrt', 'log2']
    }
    rf = sk_rfreg( random_state=0)
    #scorer = make_scorer(r2_score,sample_weight=weights)
    rs = RandomizedSearchCV(estimator=rf, param_distributions=param_grid, n_iter=15, 
                           scoring = 'neg_mean_absolute_error',
                           n_jobs=5, cv=3)
    
    rs = rs.fit(X_train, y_train)
    print('Best R^2 score', rs.best_score_)
    print ('Best Params:')
    print(rs.best_params_)

I'm running this code on an EC2 server with 128 GB RAM. RandomSearchCV spawns more than 5 python processes which slowly consume all the RAM and run out of memory before the execution completes. Here's the output of the 'top' command

Could this issue be reopened and investigated?

@rishabhgit I don't think this issue is relevant...