Predict over text not present in the DB

[rafacarrascosa]

It would be great to provide a functionality to predict a relation over some text without asking you to insert that text in to the database.
This could be really useful for applications where IEPY is mounted as a service and the corpora is not known in full a-priori.

[jmansilla]

+1

[iScienceLuvr]

Has there been any progress on this @rafacarrascosa @jmansilla ?

[rafacarrascosa]

@iScienceLuvr sorry, nothing on this yet

[iScienceLuvr]

@rafacarrascosa what other research has been done on this?

[rafacarrascosa]

@iScienceLuvr nothing at all... the issue with this issue is that there are many queries to the DB here and there that need to be 'tapped' somehow to work with something that is now in the DB. if you want the ticket it's all yours!

[iScienceLuvr]

@rafacarrascosa so how can I do this project? Idk where to start

[rafacarrascosa]

@iScienceLuvr sorry for the absurd delay.
If I were to take this issue I would start here https://github.com/machinalis/iepy/blob/develop/iepy/extraction/active_learning_core.py trying to make the predict method to work without candidates in the database...

[jmansilla]

Actually, if I recall correctly, that method is already working under the assumption that evidences are not in the DB. The challenge would be to build in your scripts Evidences resembling those that are created from the DB.

[iScienceLuvr]

@rafacarrascosa @jmansilla I do not undersrand how this code works... So predict() is the function that learns to classify? And what are the questions the are asked to the user?