Skip to content

Is there a plan to implement Spark Connect to Joblib-spark? #50

Open
@Kurdzik

Description

@Kurdzik

Hi,
Recently i've finished setting up a Spark cluster on couple of separate VMs.

When i was trying to perform SKlearn model training using Joblib-spark i've encountered following problem:

from pyspark.sql import SparkSession

spark = SparkSession.builder.remote('sc://<master node ip>').appName("JoblibSparkBackend").getOrCreate()
register_spark()

param_distributions = {
    "n_estimators": list(range(100, 500, 50)),
    "max_depth": list(range(2, 7)),
}

model = RandomForestRegressor()
random_forest = RandomizedSearchCV(model,param_distributions,cv=5,refit=True)

with parallel_backend('spark', n_jobs=4):
    random_forest.fit(X=X_train,y=y_train)
...
NotImplementedError: sparkContext() is not implemented.

Is there a workaround for this issue? Or is this something that will be implemented in a future ?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions