predictionio-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pat Ferrel <>
Subject Re: LLR and other similarity metrics graph in The Universal Recommender with CCO slide
Date Thu, 04 May 2017 16:03:27 GMT
That was generated using the old Mahout Mapreduce recommenders, which had pluggable similarity
metrics. I ran it on a vey large E-Commerce dataset from a real ecom site. The data was for
6 months of sales. We did cross-validation of an 80 training set and 20% held out probe/test
set. The test set was 20% of the most recent sales. We then measure MAP@k for several k. A
decline in MAP@k as K increases means the ranking of items is correct. This higher MAP@k the
better the precision of recommendations.

Using cross-validation between different algorithms is highly suspect so this was using an
identical algo, but not one I’d use today.

On May 4, 2017, at 8:54 AM, Marius Rabenarivo <> wrote:


Can you point me to some resource explaining how the graphic comparing
LLR with other similarity metrics was generated?



You received this message because you are subscribed to the Google Groups "actionml-user"
To unsubscribe from this group and stop receiving emails from it, send an email to
To post to this group, send email to <>.
To view this discussion on the web visit
For more options, visit <>.

View raw message