On Sat, May 5, 2012 at 12:06 AM, hao wang wrote:
>
> 1) is there anyway we can dump the weights of the features from a
> trained-model?
>
Yes. Use the model dissector or just grab the weights out of the model.
You can also access the weights matrix directly using getBeta()
> 2) in terms of the cold start, is there anyway we can apply the prior
> probability to the classifier?
>
A bi-exponential prior is built-in. You can set the weights by using
getBeta().assign(newValues). Unless you set the learning weight decay to
be slower than usual, however, this may not have much effect. This
initialization is very different from a prior distribution, though.
If you could say more about your goals, we could comment better.