Why are .csv files not directly supported with AutoML?

Question

Hello,
For the Azure Automated ML, I wanted to follow a tutorial that used the Microsoft bike rentals dataset. In the tutorial, the .csv file was uploaded directly, and the Automated ML pipeline worked fine.
When I tried using the uploaded .csv file, it was identified as the data asset is not supported.
I eventually figured out how to modify the .csv file to create an MLTable object with the requisite YAML file, and to upload the folder with the .csv and .yaml files.

Is there anyway to use a .csv file directly?

Answer 1

Hi @Brian Van Steen,

Thank you for reaching out to Microsoft Q&A forum!

Azure Automated ML supports various data formats, including CSV. However, it is possible that the CSV file you uploaded was not in the correct format or structure required by Automated ML.

I tried to repro your issue and able to use .csv files by selecting the data "type" as "Tabular":

To use a CSV file directly, ensure that it meets the following requirements:

1. The first row of the CSV file should contain the column headers.
2. The CSV file should not contain any blank rows or columns.
3. The CSV file should not contain any special characters or unsupported data types.

If your CSV file meets these requirements and is still not supported by Automated ML, you can try converting it to a supported format like Parquet or TSV.

Alternatively, you can create an MLTable object with the requisite YAML file, as you did before, and upload the folder with the CSV and YAML files. This approach allows you to customize the data preprocessing steps and transformations before training the model.

You can refer to the Azure documentation: Create and manage data assets and how to create an MLTable object for more information.

I hope this helps! Thank you.