Connector for SQL / NoSQL Databases

[sanketsarang]

Currently, the train function accepts data from the file system, URL or using a Pandas DataFrame. What would you recommend to extend support to popular databases? Use PySpark, or something else? Looking for suggestions on the basis of:

1. Support for a wide range of Databases
2. Same connector to support NoSQL databases (optional, but preferred)
3. Connector performance. Ability to read large data in short periods of time
4. Ability to handle very large data (tables with > 100M rows)
5. SQL query / data filtering ability