[Networkit] personalized PageRank implementation

Karim Kouki kookstrash at gmail.com
Wed May 10 09:31:50 CEST 2017


Hi,
Thank you again for being curious about my issue.

Let me be more clear,

for some reason I was convinced that there was some functionality in
NetworKit that could do for my challenge(see below).



Actually, the background task is binary classification.

I want to train a model to be able to make it learn the decision boundary
on a whole new unseen dataset.



To do so, I need to refine my dataset with "relational data" deduced from
the PageRank. This "relational data" consists of the influence of the class
1 throughout the network.
The problem I was experiencing had to do with the fact that I need to
import the class attribution in the first place and I didn't know how to do
that in an optimized way.


Theoretically, there would be no problem, I could just import a list of
nodes. But I have to think of a scaling strategy. Indeed, the data I am
processing is potentially very large and could be ever growing. This being
the main reason why I have to think in optimization terms.




The way I see it now is that I just need to sort my data by node Ids. The
first subset would just be the '0' and the last part would be the '1'. The
separating id is then the only information that is needed to  get the class
attribution information without having to import the whole node list and
check whether each node belongs or not to the list of nodes, in my opinion.
Thanks to this trick there only remains to check if the current node id is
below the threshold id or not to know which class it belongs to.


Kind regards,
Karim



2017-05-09 19:21 GMT+02:00 Christian Staudt <christian.staudt at kit.edu>:

>
> On 9. May 2017, at 17:32, Karim Kouki <kookstrash at gmail.com> wrote:
>
>
> Nevertheless, about the teleport term, I need to specify a subset of nodes
> that will receive the additional teleport term as contribution of special
> nodes (fraudulent transaction nodes).
>
>
> My problem is that I don't know what to do for that specifically. I found
> two objects in the documentation: NetworKit::Partition and NetworKit::Cover
> that seem to do the trick but this implies I need to read a supplementary
> text file (in edgelist format) to load a Partion/Cover in memory and check,
> for each node in the loop, whether or not it belongs to a partition/Cover
> and add the teleportation term in the positive alternative.
>
>
> Hi Karim,
> maybe I don’t get the problem, but maybe there is no problem: Why can’t a
> simple list with the nodes in the subset be used?
>
> Best regards
> Chris
>
>
>
> _______________________________________________
> NetworKit mailing list
> NetworKit at ira.uni-karlsruhe.de
> https://lists.ira.uni-karlsruhe.de/mailman/listinfo/networkit
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ira.uni-karlsruhe.de/pipermail/networkit/attachments/20170510/c7638a96/attachment.html>


More information about the NetworKit mailing list