You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@mahout.apache.org by Lance Norskog <go...@gmail.com> on 2010/09/01 00:55:10 UTC
Re: SVDRecommender convergence?
About the SVDRecommender- 10 features and 50 iterations gave
evaluation scores from .70 to .81, solely by changing the random
seeds.
On Tue, Aug 31, 2010 at 12:13 AM, Sean Owen <sr...@gmail.com> wrote:
> I don't have any good rules of thumb for you -- maybe the author can chime in.
> It should be a fairly standard implementation and I would not expect
> unusual behavior in this regard, but can't say I know either way.
>
> But are you asking a question about the recommender or the evaluator?
>
> On Tue, Aug 31, 2010 at 7:35 AM, Lance Norskog <go...@gmail.com> wrote:
>> Hi-
>>
>> How many features and how many iterations should make the
>> SVDRecommender converge?
>>
>> I used the GroupLens example and 10k dataset with the SVDRecommender
>> instead of the SlopeOneRecommender. The
>> AverageAbsoluteDifferenceRecommenderEvaluator is very sensitive to the
>> random seed. It's on my laptop, so experimenting with # of features
>> and # of iterations got impossible really fast.
>>
>> --
>> Lance Norskog
>> goksron@gmail.com
>>
>
--
Lance Norskog
goksron@gmail.com
Re: SVDRecommender convergence?
Posted by Sean Owen <sr...@gmail.com>.
Mahout in Action? Yes it's in there. The output is the average
absolute difference between the actual and estimated rating. GroupLens
ratings are from 1 to 5. 0.7 average error isn't bad on that scale.
On Thu, Sep 2, 2010 at 3:26 AM, Lance Norskog <go...@gmail.com> wrote:
> Ah! What exactly does 0.70 mean as an evaluator output?
>
> If this is not in HiAction, that's an idea... I don't have my MEAP 0.3 handy.
Re: SVDRecommender convergence?
Posted by Lance Norskog <go...@gmail.com>.
Ah! What exactly does 0.70 mean as an evaluator output?
If this is not in HiAction, that's an idea... I don't have my MEAP 0.3 handy.
On Wed, Sep 1, 2010 at 1:07 AM, Sean Owen <sr...@gmail.com> wrote:
> All in all that sounds roughly reasonable to me -- that's a reasonable
> number of features and iterations, and the eval result is varying by
> just about 3% (scale of 1-5).
>
> On Tue, Aug 31, 2010 at 11:55 PM, Lance Norskog <go...@gmail.com> wrote:
>> About the SVDRecommender- 10 features and 50 iterations gave
>> evaluation scores from .70 to .81, solely by changing the random
>> seeds.
>
--
Lance Norskog
goksron@gmail.com
Re: SVDRecommender convergence?
Posted by Sean Owen <sr...@gmail.com>.
All in all that sounds roughly reasonable to me -- that's a reasonable
number of features and iterations, and the eval result is varying by
just about 3% (scale of 1-5).
On Tue, Aug 31, 2010 at 11:55 PM, Lance Norskog <go...@gmail.com> wrote:
> About the SVDRecommender- 10 features and 50 iterations gave
> evaluation scores from .70 to .81, solely by changing the random
> seeds.