active learning

[JL Meunier]

Hi Andreas,

I'm wondering: to what extend can we use the warm_start mechanism to do active learning?

The idea would be:

- calling fit on some train dataset

- predict for some additional data and have a user to review those predictions and fix if needed

- warm start, and train again with the initial train dataset extended with the additional annotated data

(Actually, having a notion of prediction confidence would be useful.)

Currently I'm training with OneSlackSSVM. I'm unsure how this algo would deal with this approach ...and your 2009 paper is pretty technical!! :-)

Thanks

JL

[Andreas Mueller]

Hi JL.
You could look at the energy assigned by the model as a measure of confidence, but that only makes sense in relative terms compared to other possible predictions.
And the warm_start will only help you if the model building is the bottle neck. If you have user interactions, the human is probably the bottleneck and you could
just retrain the whole model while they are clicking around.

Andy

--
You received this message because you are subscribed to the Google Groups "pystruct" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pystruct+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.