mahout preferences range

[Zia mel <zi...@gmail.com>]

Hi

Q1:  If I have preferences that range from -1 to -5 with higher -1 and
lower -5 is that fine or I need to make the preferences positive ?

Q2: Does the range of preferences matter ? for examples is there
difference if the preferences are between 0 to 1 or 0 to 5 ? According
to a test I have done it seems yes, is that correct? Is there a
prefered range of preferences ?

Thanks !

Re: mahout preferences range

[Zia mel <zi...@gmail.com>]

Yes for the recommender . I had a range from 0 to 1 and I mapped it
using log which gives more range but in negative and this gives better
precision results.

On Sat, Dec 15, 2012 at 9:33 AM, Sean Owen <sr...@gmail.com> wrote:
> I assume you are talking about recommenders. For most everything you
> see, range does not matter in the sense that there is not a better or
> worse range to use.
>
> It will affect some results slightly. For example, doubling all your
> numbers should double your RMSE evaluation scores of course. Shifting
> your range would change an  uncentered cosine similarity measure.
>
> Finally note that most approaches involve random sampling somewhere.
> This can make for different results on different runs too.
>
> On Sat, Dec 15, 2012 at 3:10 PM, Zia mel <zi...@gmail.com> wrote:
>> Hi
>>
>> Q1:  If I have preferences that range from -1 to -5 with higher -1 and
>> lower -5 is that fine or I need to make the preferences positive ?
>>
>> Q2: Does the range of preferences matter ? for examples is there
>> difference if the preferences are between 0 to 1 or 0 to 5 ? According
>> to a test I have done it seems yes, is that correct? Is there a
>> prefered range of preferences ?
>>
>> Thanks !

Re: mahout preferences range

[Sean Owen <sr...@gmail.com>]

I assume you are talking about recommenders. For most everything you
see, range does not matter in the sense that there is not a better or
worse range to use.

It will affect some results slightly. For example, doubling all your
numbers should double your RMSE evaluation scores of course. Shifting
your range would change an  uncentered cosine similarity measure.

Finally note that most approaches involve random sampling somewhere.
This can make for different results on different runs too.

On Sat, Dec 15, 2012 at 3:10 PM, Zia mel <zi...@gmail.com> wrote:
> Hi
>
> Q1:  If I have preferences that range from -1 to -5 with higher -1 and
> lower -5 is that fine or I need to make the preferences positive ?
>
> Q2: Does the range of preferences matter ? for examples is there
> difference if the preferences are between 0 to 1 or 0 to 5 ? According
> to a test I have done it seems yes, is that correct? Is there a
> prefered range of preferences ?
>
> Thanks !