Default weights

[Omid]

Hi, 

I am using the CRF, I pass my custom wordfeaturizer to it. when it starts to run I can see following message in the terminal:

[main] INFO epic.sequences.CRFModel - Reading old weights from epic.sequences.CRFModel.weights.txt.gz

It looks that I'm doing something incorrect that it want to load some "old weights". Or this message is always there?

[David Hall]

Mostly for debugging, I cache the weights in a gzipped text file during training. It tries to read them in. Nothing goes wrong if they're not there.

[Omid]

Is there any way to turn it off? It seems it cause this:

[main] INFO epic.sequences.TaggedSequenceModelFactory - NumFeatures 81/81: 1187 (0.046s elapsed.)
[main] INFO epic.sequences.TaggedSequenceModelFactory - There are 37 and 2328 features total.

[main] INFO epic.sequences.CRFModel - Reading old weights from epic.sequences.CRFModel.weights.txt.gz

Exception in thread "main" java.lang.OutOfMemoryError: GC overhead limit exceeded

I didn't get this exception before, I even revert my code to a version that was working and I still get this exception, I really don't know what is the problem!

[Omid]

my problem solved by deleting epic.sequences.CRFModel.weights.txt.gz I don't know why it had been created in first place.

[David Hall]

Huh, it really shouldn't use that much memory. I'll look into disabling it.