LinearLearner local mode - ShardedByS3Key is not currently supported in Local Mode

[russau]

Please fill out the form below.

System Information

• Framework (e.g. TensorFlow) / Algorithm (e.g. KMeans): LinearLearner
• Framework Version: sagemaker==1.11.0
• Python Version: 3.6.5
• CPU or GPU: CPU
• Python SDK Version: sagemaker==1.11.0
• Are you using a custom image: no

Describe the problem

Attempting to use the LinearLearner estimator in local mode throws the exception:

RuntimeError: DataDistribution: ShardedByS3Key is not currently supported in Local Mode

Minimal repro / logs

To repro I'm just taking the multiclass classifiers example and switching it use to local mode.

multiclass_estimator = sagemaker.LinearLearner(role=sagemaker.get_execution_role(),
                                               train_instance_count=1,
                                               train_instance_type='local',
                                               predictor_type='multiclass_classifier',
                                               num_classes=7)

Digging thru the code I see that amazon_estimator.py sets the s3_input to distribution='ShardedByS3Key'.

Is there a way to create a FullyReplicated recordset and get local mode working?

[iquintero]

Hi @russau

Linear Learner is not supported in Local Mode (none of the Amazon 1P Algorithms are). These containers are not public and can't be pulled into your local environment.

[laurenyu]

Added documentation in #575 - going to close this issue now. Feel free to reopen if necessary.