Technology behind AutoML support in the Platform

For AutoML support we have used MindsDB. MindsDB is a technology that allows you to create, train and deploy Machine Learning models on different databases, abstracting the models as AI Tables. Besides, the predictions will be available through SQL queries.

By bringing machine learning into the database, which is the usual place to keep the data, MindsDB accelerates the speed of machine learning development.

Its main features include:

Automatic data pre-processing, feature engineering and coding.
Classification, regression, and time series tasks.
Bring models into production without "traditional deployment" like AI Tables.
Get the accuracy of the models.
Join ML models with existing data.
Anomaly detection.
Explainability analysis of the models.
GPU support for model training.
Open JSON-AI syntax to build models and to provide ML blocks declaratively.

How does it work?

Connect MindsDB with the database on which you will create the models. MindsDB supports different database and persistence technologies:

Train a Predictor using a single SQL statement (MindsDB will learn from the historical data automatically) or import your own ML model into a Predictor via JSON-AI:

Let's assume the table income_table, for which you want to predict the debt attribute based on income:

Without a model you can't predict the debt for an income that you don't have in the table:

By creating a PREDICTOR on this table, MindsDB will create a linear regression to predict debt:

Make predictions with SQL statements (The Predictor is exposed as AI virtual tables). There is no need to deploy the models as they are part of the data layer. You can now query your model:

MindsDB is capable of choosing the most suitable model based on the input attributes, as well as selecting a percentage of the input data to validate the model and give the accuracy of the model: