ML2SQL: Compiling a declarative machine learning language to SQL and python

This demonstration presents a machine learning language MLearn that allows declarative programming of machine learning tasks similarly to SQL. Our demonstrated machine learning language is independent of the underlying platform and can be translated into SQL and Python as target platforms. As modern hardware allows database systems to perform more computational intense tasks than just retrieving data, we introduce the ML2SQL compiler to translate machine learning tasks into stored procedures intended to run inside database servers running PostgreSQL or HyPer. We therefore extend both database systems by a gradient descent optimiser and tensor algebra. In our evaluation section, we illustrate the claim of running machine learning tasks independently of the target platform by comparing the run-time of three in MLearn specified tasks on two different database systems as well as in Python. We infer potentials for database systems on optimising tensor data types, whereas database systems show competitive performance when performing gradient descent.