AzureML HyperDriveStep is incorrectly reading Input port names as duplicates

Question

I'm trying to build a regression model pipeline using a HyperDriveStep for parameter tuning.

When I pass in my features dataset and targets dataset as named inputs, if I run this in a notebook with a fresh kernel, I get an error that says ValueError: [features_data] is repeated. Input port names must be unique. (See first screenshot)

If I comment out the features data input and re-run the cell, the HyperDriveStep builds with no issue. (But obviously I can't run the step and train my model without the features.) (See second screenshot)

If I un-comment the features data input and re-run the cell, this time the HyperDriveStep builds with no issue, and I can run my pipeline. (See third screenshot)

Error report is as follows:

---

ValueError Traceback (most recent call last)
Input In [7], in <cell line: 25>()
21 # Risk: need to ensure tha the model_file entered here is the same name
22 # as what's saved in training/train_{model_algorithm}_regressor.py
24 hd_step_name=f'hd_training_step_{segment_type}_{model_algorithm}'
---> 25 hd_step = HyperDriveStep(
26 name=hd_step_name,
27 hyperdrive_config=hd_config,
28 inputs=[
29 training_features.as_named_input('features_data'),
30 training_targets.as_named_input('targets_data')],
31 outputs=[metrics_data, saved_model],
32 allow_reuse=False)

File /anaconda/envs/azureml_py38/lib/python3.8/site-packages/azureml/pipeline/steps/hyper_drive_step.py:249, in HyperDriveStep.init(self, name, hyperdrive_config, estimator_entry_script_arguments, inputs, outputs, metrics_output, allow_reuse, version)
246 self._params[HyperDriveStep._primary_metric_goal] = hyperdrive_config._primary_metric_config['goal'].lower()
247 self._params[HyperDriveStep._primary_metric_name] = hyperdrive_config._primary_metric_config['name']
--> 249 super(HyperDriveStep, self).init(name=name, inputs=inputs, outputs=outputs,
250 arguments=estimator_entry_script_arguments)

File /anaconda/envs/azureml_py38/lib/python3.8/site-packages/azureml/pipeline/core/builder.py:149, in PipelineStep.init(self, name, inputs, outputs, arguments, fix_port_name_collisions, resource_inputs)
146 resource_input_port_names = [PipelineStep._get_input_port_name(input) for input in resource_inputs]
147 output_port_names = [PipelineStep._get_output_port_name(output) for output in outputs]
--> 149 PipelineStep._assert_valid_port_names(input_port_names + resource_input_port_names,
150 output_port_names, fix_port_name_collisions)
152 self._inputs = inputs
153 self._resource_inputs = resource_inputs

File /anaconda/envs/azureml_py38/lib/python3.8/site-packages/azureml/pipeline/core/builder.py:190, in PipelineStep._assert_valid_port_names(input_port_names, output_port_names, fix_port_name_collisions)
188 if input_port_names is not None:
189 assert_valid_port_names(input_port_names, 'input')
--> 190 assert_unique_port_names(input_port_names, 'input')
192 if output_port_names is not None:
193 assert_valid_port_names(output_port_names, 'output')

File /anaconda/envs/azureml_py38/lib/python3.8/site-packages/azureml/pipeline/core/builder.py:184, in PipelineStep._assert_valid_port_names.<locals>.assert_unique_port_names(port_names, port_type, seen)
182 for port_name in port_names:
183 if port_name in seen:
--> 184 raise ValueError("[{port_name}] is repeated. {port_type} port names must be unique."
185 .format(port_name=port_name, port_type=port_type.capitalize()))
186 seen.add(port_name)

ValueError: [features_data] is repeated. Input port names must be unique.

Answer 1

@Cliff Clive Thanks for the question. Can you please add more details about the sample and SDK version that you are using.

HyperDrive is an automated hyperparameter tuning system ​that helps users find better results in a cost effective manner.