Redesigning the software
Here we can gather ideas to get the software to be nicer (since it's quite hackish as it is)
Requirements
Online Learning Use Cases
- Sequential Access to a list of databases
- Binary Database Operations (cost function, dot product)
- Overlay databases
- Convert feature query to collection query and counting query.
I2 Use cases
- Scored Select Query
- unify table (make distinct)
- Select rows of table A which are not in table B
- Select rows which are active in a boolean vector
- Use result of a query as values in other query
- Use Tables that contain all elements of a namespace
- Interpret several result columns as representing rows in certain tables
Architecture
- Lets try to adapt the DBMS architecture and add a few special operations to the layers to take account of solving operations
- tables should have integer and double columns
- User Layer (Model, Constraint, etc. all proper convenient classes from a client perspective)
- Processing Layer (mapping of the above into arrays, ints, doubles, optimized for fast processing)
Random Requests
- Query Should have a target types (i.e. you can define which tables the select columns represent)
- Table "closing" etc. should be more transparent.
--
SebastianRiedel - 26 Oct 2006