Why does Ciaos dataset with PMF get RMSE 2.823082? about librec

[guoguibing]

Better performance may be obtained after tuning the moment larger.

[asoboy]

Thanks for your commenting!!

The Ciao dataset has 35835 ratings and is downloaded from here (http://www.public.asu.edu/~jtang20/datasetcode/ciao_with_rating_timestamp_txt.zip).

I tested with momentum from 0.1 to 0.9 (the 1.0 will cause crach due to the Loss grows into infinity), and the result is here:

learning Rate = 50, regP = regQ = 0.1, K = 10, iters = 100

      MAE     RMSE    Momentum

PMF 2.484504 2.807264 0.1
PMF 2.516064 2.832653 0.2
PMF 2.471362 2.803306 0.3
PMF 2.452357 2.783117 0.4
PMF 2.464851 2.797973 0.5
PMF 2.582281 2.890350 0.6
PMF 2.562090 2.872376 0.7
PMF 2.591306 2.898688 0.8
PMF 2.553969 2.868095 0.9

I am not sure about whether it is the dataset's flaw or not. Could you help run with this dataset and return the confiuguration and result of PMF, Thank you!!

[guoguibing]

moment is not restricted in [0, 1], try larger values like 10, 50, etc.

I do not have spare time to run your tests.

[guoguibing]

BTW, one of my collaborators mentioned that you may want to check the format of ciao data set.