Re: EM - distributionForInstance question

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Re: EM - distributionForInstance question

Eibe Frank
Are you working with high-dimensional data? That might be why you are
getting these discrete probabilities.

Cheers,
Eibe

On May 6, 2005, at 6:40 AM, Daniel Stephan wrote:

> Just a quick one to be sure:
>
> The method distributionForInstance gives an array of doubles. Currently
> it is the case that for every instance exactly one of the doubles is
> 1.0
> and all the others are 0.0. Is that expected?
>
> It is not always the same, though, every index(cluster) gets its fair
> share of instances for which its value is 1.0. I could now infer a hard
> cluster assignment where every instance is assigned to exactly one
> cluster (only one will have the 1.0 for that instance). I just THOUGHT
> it would give a more fuzzy result.
>
> Is it normal?
>
> Cheers
> Daniel
>
>
> _______________________________________________
> Wekalist mailing list
> [hidden email]
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist


_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Reply | Threaded
Open this post in threaded view
|

Re: EM - distributionForInstance question

Daniel Stephan
Uhmm... I have 50 to 300 dimensions (already compressed from about 2000 to
20000), depending on my setup, every value being in the range -1 to +1.
Is that too much?

I mean, I can't just leave out dimensions. Is there anything else I can do?

Cheers,
Daniel



Eibe Frank schrieb:

> Are you working with high-dimensional data? That might be why you are
> getting these discrete probabilities.
>
> Cheers,
> Eibe
>
> On May 6, 2005, at 6:40 AM, Daniel Stephan wrote:
>
>> Just a quick one to be sure:
>>
>> The method distributionForInstance gives an array of doubles. Currently
>> it is the case that for every instance exactly one of the doubles is 1.0
>> and all the others are 0.0. Is that expected?
>>
>> It is not always the same, though, every index(cluster) gets its fair
>> share of instances for which its value is 1.0. I could now infer a hard
>> cluster assignment where every instance is assigned to exactly one
>> cluster (only one will have the 1.0 for that instance). I just THOUGHT
>> it would give a more fuzzy result.
>>
>> Is it normal?
>>
>> Cheers
>> Daniel
>>
>>
>> _______________________________________________
>> Wekalist mailing list
>> [hidden email]
>> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>
>
>


_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist