Error on Azure ML - AutoMLPipelineBuilder

Question

I want to build a Pipeline in Azure ML. The Training pipeline runs well. Training:

training_pipeline_steps = AutoMLPipelineBuilder.get_many_models_train_steps(
    experiment=experiment,
    train_data=full_dataset,
    compute_target=compute_target,
    node_count=1,
    process_count_per_node=4,
    train_pipeline_parameters=hts_parameters,
    run_invocation_timeout=3900,
)

Forecasting:

from azureml.train.automl.runtime._hts.hts_parameters import HTSInferenceParameters

inference_parameters = HTSInferenceParameters(
    hierarchy_forecast_level="Material",  # The setting is specific to this dataset and should be changed based on your dataset.
    allocation_method="proportions_of_historical_average",
)

steps = AutoMLPipelineBuilder.get_many_models_batch_inference_steps(
    experiment=experiment,
    #inference_data=registered_inference,
    inference_data = full_dataset,
    compute_target=compute_target,
    inference_pipeline_parameters=inference_parameters,
    node_count=1,
    process_count_per_node=4,
    arguments=["--forecast_quantiles", 0.1, 0.9],
)

But I always get the error on Forecasting:

Can anybody help with that error? Thank you!

Answer 1

I have tried your code in my environment with different dataset, and it worked successfully.

Prediction

So, configuration for batch inference will be wrong or not appropriate for your data, check that once. And also check the detailed logs in output+logs -> user -> stdout -> 0 -> process000.std.txt.

Answer 2

I had the same problem and lately solved.

Make sure the inference pipeline is in the same experiment than the training pipeline. If not, you have to specify the experiment name with the 'train_experiment_name' parameter in the get_many_models_batch_inference_steps arguments.

hope it helps.