Fwd: Reg PFP Growth Algorithm

[cvkkumar <cv...@me.com>]

Hi,

I am new to Mahout and was going through the source code of PFP Growth to understand it. I got confused at a point before the final aggregation of results, because I could not understand how it could avoid double counting frequent patterns from different groups.

For instance, if we consider this scenario where in 

X-Y-Z  is a branch that fall in Group No 2. (because Z is in Group 2)
X-Y is a branch that falls in Group No 1. (because Y is in Group 1)

Let all the combinations of X Y Z be frequent. 

Now, I dont understand if  X-Y would be counted as frequent patterns from both these groups. Intuitively, from the PFP Growth paper, I thought it should be returned as a frequent pattern only from Group 1. 

Am I correct? Is there something that I am missing. I would be grateful if someone points it out!;

Thanks in advance!
Regards,
Krishna.

Re: Reg PFP Growth Algorithm

[Robin Anil <ro...@gmail.com>]

On Mon, Dec 6, 2010 at 9:19 AM, cvkkumar <cv...@me.com> wrote:

>
>
> Hi,
>
> I am new to Mahout and was going through the source code of PFP Growth to
> understand it. I got confused at a point before the final aggregation of
> results, because I could not understand how it could avoid double counting
> frequent patterns from different groups.
>
> For instance, if we consider this scenario where in
>
> X-Y-Z  is a branch that fall in Group No 2. (because Z is in Group 2)
> X-Y is a branch that falls in Group No 1. (because Y is in Group 1)
>
> Let all the combinations of X Y Z be frequent.
>
> Now, I dont understand if  X-Y would be counted as frequent patterns from
> both these groups. Intuitively, from the PFP Growth paper, I thought it
> should be returned as a frequent pattern only from Group 1.
>
> The patterns may not be of the same frequency, so XY from group 1 will
override the count of XY generated from group 2. But XYZ will exist only in
group 2. (thats how the division is done). So there is no problem of merging
it


> Am I correct? Is there something that I am missing. I would be grateful if
> someone points it out!;
>
> Thanks in advance!
> Regards,
> Krishna.
>
>