itemcf3 in GraphChi takes too long

[Hao Wang]

I'm playing with itemcf3 in GraphChi toolkits and it's like taking forever to compute a user-item rating matrix containing 30 million nozeros in the matrix. Anybody has any idea what might be going wrong? I'm using the following command to run the code:

toolkits/collaborative_filtering/itemcf3 --training=graphchi_input.txt --K 20 --quiet=1

[Danny Bickson]

Hi Hao, 

itemcf3 is an experimental code. You should be using itemcf/itemcf2.

The difference is in the implemented cost functions. Those are the supported cost functions:

FOR itemcf:  --distance=XX, 0 = Jaccard index, 1=AA, 2=RA, 3=Aiolli
FOR itemcf2: --distance=XX, 3 = PEARSON, 4=COSINE, 5=CHEBYCHEV, 6=MANHATTEN, 7=TANIMOTO, 8=LOG_LIKELIHOOD, 9 = SLOPE_ONE

Running time is mainly affected from the number of user and item pair and less from the number of non zeros... How many users/items do you have?

Best,

		Danny Bickson Co-Founder US phone: 206-691-8266 Israeli phone: 073-7312889

On Wed, May 7, 2014 at 6:12 AM, Hao Wang <hao...@gmail.com> wrote:

I'm playing with itemcf3 in GraphChi toolkits and it's like taking forever to compute a user-item rating matrix containing 30 million nozeros in the matrix. Anybody has any idea what might be going wrong? I'm using the following command to run the code:

toolkits/collaborative_filtering/itemcf3 --training=graphchi_input.txt --K 20 --quiet=1

--
You received this message because you are subscribed to the Google Groups "graphchi-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to graphchi-discu...@googlegroups.com.
To post to this group, send email to graphchi...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/graphchi-discuss/f07ff2ae-6013-4d24-9380-c9b39e42d1b6%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Hao Wang]

Thank you for your response. We're playing with itemcf2 now and we'll see how things are going.

The following is the header of our input file. We have ~25 million users and ~2 million items.

%%MatrixMarket matrix coordinate real general

% Generated 22-Apr-2014

24654148        1991459 35145915

--
You received this message because you are subscribed to a topic in the Google Groups "graphchi-discuss" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/graphchi-discuss/d75mfzJyUp8/unsubscribe.
To unsubscribe from this group and all its topics, send an email to graphchi-discu...@googlegroups.com.

To post to this group, send email to graphchi...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/graphchi-discuss/CAGKFuX9o0MuC6DHWOG2hidy68pX6YgEChYS_09JD48hkE_VYig%40mail.gmail.com.

[Hao Wang]

Dear Danny,

I invented a recommender system for fair AI that not only improves technical accuracy but also fairness metrics. Are you interested in integrating the algorithm into your toolkit ?

Best Regards,

Hao Wang

You received this message because you are subscribed to a topic in the Google Groups "graphchi-discuss" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/graphchi-discuss/d75mfzJyUp8/unsubscribe.
To unsubscribe from this group and all its topics, send an email to graphchi-discu...@googlegroups.com.

To post to this group, send email to graphchi...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/graphchi-discuss/CAGKFuX9o0MuC6DHWOG2hidy68pX6YgEChYS_09JD48hkE_VYig%40mail.gmail.com.