Eval in epic

[Omid]

I find that in Epic for running the test set for CRF, we should run this:

val stats = TaggedSequenceEval.eval(crf, test)

My question is where are the exact output of CRF on input and How can I access it?

Thanks

[David Hall]

i don't follow?

[Omid]

My question is how can I get the output of CRF on the test set:

In your example you used: val stats = TaggedSequenceEval.eval(crf, test)

are the output labels in "stats" ?

[David Hall]

oh, you mean the actual guesses? You have to do that separately, sadly. test.map(ts => crf.bestSequence(ts.words).render foreach println or something like that

[Omid]

Ahan, Okay, thanks ...

what is the default learning algorithm for your CRF implementation?  

[David Hall]

are you using buildSimple? There's an OptParams object that has a few options in it (regularization, useStochastic, useL1) that control the regularization and kind of gradient minimizer used. The objective itself is log loss.

[Omid]

Yes, I'm using buildSimple