how do I output the predict results only in the testset

Question

how do I output the predict results only in the testset

danyang-liu opened this issue 6 years ago · 11 comments

I use BPR. I want to get the predict results for the u-i pairs in testset. But the results in ./result/ is the top_n item_rating for these users, not the items I want to predict.

Answer 1 · 2018-10-19T09:09:02.000Z

After training, You can run the predict(int userIdx, itemIdx) function with the testMatrix.

    // after training
    for (MatrixEntry me : testMatrix) {
        int userIdx = me.row(); // user
        int itemIdx = me.column(); // item
        double predictRating = predict(userIdx, itemIdx);
    }

Answer 2 · 2018-10-21T14:19:24.000Z

Can predict() be run form command line?

Answer 3 · 2018-10-22T08:41:24.000Z

Hi SunYatong:
Thanks, I will try.

Answer 4 · 2019-02-08T12:10:27.000Z

Can we predict values for a test set from the command line?

Answer 5 · 2019-02-08T12:13:23.000Z

Can we predict values for a test set from the command line? That is, instead of storing predicted values for training data or top-N values among the whole item set (in case of ranking based recommendation algorithm), can we store predicted values of a given test set from command line?

Answer 6 · 2019-02-11T01:29:52.000Z

@mvijaikumar Command line doesn't support this. You can denote the test set before training with TestSetSplitter, and add the code after training.

    // after training
    for (MatrixEntry me : testMatrix) {
        int userIdx = me.row(); // user
        int itemIdx = me.column(); // item
        double predictRating = predict(userIdx, itemIdx);
    }

Answer 7 · 2019-02-11T11:28:32.000Z

@SunYatong Thank you for your response. I am facing a peculiar problem in doing so as described above. If (userid, itemid) pair is not a part of a training set, and part of a test set (for obvious reason), testMatrix does not contain that pair.

I tested this on PMFRecommender. I could not understand why it is happening. Can you help me out?

Answer 8 · 2019-02-11T12:15:47.000Z

So you want to get the prediction of arbitrary user-item pairs? @mvijaikumar

Answer 9 · 2019-02-11T12:22:36.000Z

I want to predict rating for user and item that is part of training set, however, the exact pair is not a part of training set. That is, user is part of training set, itme is part of training set.

…

On Mon, Feb 11, 2019, 17:45 SunYatong ***@***.*** wrote: So you want to get the prediction of arbitrary user-item pairs? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#280 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AeROdZSWIathgc6vjU4ZVqZ5Ud9tg48Oks5vMV74gaJpZM4XbJ9G> .

Answer 10 · 2019-02-11T12:28:11.000Z

I want to predict rating for user and item that is part of training set, however, the exact pair is not a part of training set. That is, user is part of training set, item is part of training set.

…

On Mon, Feb 11, 2019, 17:45 SunYatong ***@***.*** wrote: So you want to get the prediction of arbitrary user-item pairs? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#280 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AeROdZSWIathgc6vjU4ZVqZ5Ud9tg48Oks5vMV74gaJpZM4XbJ9G> .

Answer 11 · 2019-02-12T01:02:36.000Z

So you do want to get the prediction of arbitrary user-item pairs... @mvijaikumar

Just use

 predict(userIdx, itemIdx)