Azure
AutoML-Image-Classification Tutorial Notebook Error: MLTable input is invalid.
Hello,
My colleague and I are having similar issues when runing your image classification and object detection tutorial notebooks. After running the notebooks and submitting the automl-image job runs, we received following errors (similar for classficaiton and detection tasks):
MLTable input is invalid. Failed with visit error: Failed with execution error: error in streaming from input data sources VisitError(ExecutionError(StreamError(NotFound))) => Failed with execution error: error in streaming from input data sources ExecutionError(StreamError(NotFound))
The classification notebook that I followed is this one: https://github.com/Azure/azureml-examples/blob/sdk-preview/sdk/jobs/automl-standalone-jobs/automl-image-classification-multiclass-task-fridge-items/automl-image-classification-multiclass-task-fridge-items.ipynb
Please help! Thanks a lot!
He
Adding to the thread. For the same notebook it would be good if you added the following line of code to cell 7. When executing the notebook, people will run into an error that the ml table directories do not exist, As such, a mkdir operation is useful.
from jsonl_converter import convert_mask_in_VOC_to_jsonl
from pathlib import Path
Path("./data/training-mltable-folder/").mkdir(parents=True, exist_ok=True)
Path("./data/validation-mltable-folder/").mkdir(parents=True, exist_ok=True)
data_path = "./data/odFridgeObjectsMask/"
convert_mask_in_VOC_to_jsonl(data_path, uri_folder_data_asset.path)
Another issue:
When trying to train specifying the remote location of the files in the workspace datastore, the following error occurs: Error: Input request is invalid .
The tutorial, as such, is currently not executable. As you see, my colleague He has the same problem.
Issue was resolved in the meantime with the release of mltable: Create Data Assets - Azure Machine Learning | Microsoft Learn.
Please refer to the following documentation to access data assets via SDK v2 and make sure you use the latest AzureML example repository.
The PR for downloading the dataset and creating MLTable on the fly on user specified location in the PR (https://github.com/Azure/azureml-examples/pull/1911) is merged. Please feel free to raise another issue if you still face the issue.
