You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@mahout.apache.org by XiaoboGu <gu...@gmail.com> on 2011/06/06 10:06:39 UTC

What's the best practice to specify df for TPrior and alphaByLambda for ElasticBandPrior

Is 0.5 a good point to start?


Re: Taste-Web

Posted by Sean Owen <sr...@gmail.com>.
In 0.5? it's there.
In HEAD, it has been merged with utils into "integration". It's all the same
thing, except that the process for building a .war automatically is gone.

On Mon, Jun 6, 2011 at 1:03 PM, majin <ma...@x-playaz.de> wrote:

> Hey,
>
> perhaps that is now embarrassing, but what happened to taste-web folder?
> couldnt find anything about it in the release notes.
>
> Thanks.
>

Taste-Web

Posted by majin <ma...@x-playaz.de>.
Hey,

perhaps that is now embarrassing, but what happened to taste-web folder? 
couldnt find anything about it in the release notes.

Thanks.

Re: What's the best practice to specify df for TPrior and alphaByLambda for ElasticBandPrior

Posted by Ted Dunning <te...@gmail.com>.
Here is the link:

http://www.stat.columbia.edu/~gelman/research/unpublished/*prior*s7.pdf


On Mon, Jun 6, 2011 at 4:39 AM, Ted Dunning <te...@gmail.com> wrote:

> This paper has good guidance.
>
> I think that their recommendation of df=1 and scale=2.5 is very reasonable.
>  Their suggestion of mean subtraction would be disastrous, but much of the
> rest is very good.
>
> df=0.5 is a very (VERY) long tailed distribution.  Even for df=1, there is
> no mean or variance.
>
> I would actually consider using df=2 as well.
>
> In practice, I have found that the L1 prior encourages sparsity more
> strongly than the T prior.  It does, however, entangle the regularization
> factor with the scale of the prior.
>
>
> On Mon, Jun 6, 2011 at 1:06 AM, XiaoboGu <gu...@gmail.com> wrote:
>
>> Is 0.5 a good point to start?
>>
>>
>

Re: What's the best practice to specify df for TPrior and alphaByLambda for ElasticBandPrior

Posted by Ted Dunning <te...@gmail.com>.
This paper has good guidance.

I think that their recommendation of df=1 and scale=2.5 is very reasonable.
 Their suggestion of mean subtraction would be disastrous, but much of the
rest is very good.

df=0.5 is a very (VERY) long tailed distribution.  Even for df=1, there is
no mean or variance.

I would actually consider using df=2 as well.

In practice, I have found that the L1 prior encourages sparsity more
strongly than the T prior.  It does, however, entangle the regularization
factor with the scale of the prior.

On Mon, Jun 6, 2011 at 1:06 AM, XiaoboGu <gu...@gmail.com> wrote:

> Is 0.5 a good point to start?
>
>