What is AutoML?
AutoML stands for Automated Machine Learning and is a tool within Fabric that compares and executes models based on the input data and the desired outcome you want from your machine learning model. This is often used in conjunction with your data lakehouse hosted within Fabric.
Behind the scenes of AutoML
- It is built off of an open library called FLAML to deliver/ execute different types of models.
- This is the main library that allows for model selection and splits your input data 80/20. With 80 % used to train the model, and 20% to validate.
Naturally, there are some caveats to this split; one of them being that you are unable to change this split. Unless you customise the code itself or select a specific model that you know has a different Data split, proficiency in Python & the models you are using would be needed to make these amendments.
What Machine Learning Tasks can AutoML facilitate?
Binary Classification
Placing data into 1 of 2 Distinct Buckets
Multi-class classification
1 of several labels to 1 record in structured data based on features within the dataset. (Like tagging).
Often used for Customer Segmentation (Assigning customers as High value depending on their demographic or regional data)
Regresssion
Wanting to estimate a specific Value based on variables that could affect it.
Forcasting
Forming predictions based on historical data.
Considerations
During exploration of AutoML, I found that the outputs of the model are a lot better when the classification is simple, and the dataset and tasks required from the model are well defined.
There was also a lot of back-and-forth between the notebook and modifying the code. And checking the script's results, which took me some time to understand. So I would highly recommend proficiency in Python when using this tool. This would let you spot and check the code the AutoML tool creates.