Closed songweige closed 7 years ago
i.e. The denominator of MAP should be the total number of relevant documents instead of the number of relevant document before truncation. Otherwise, for example, we have document list(1, 0, 0, 0, 0, 0, 1). If we swap i=5 and i=7, then the MAP@5 will reduce from 1 to 0.7.
Hi Songwei, it looks like you are correct. Thanks for catching this!
Tests are failing due to one test case [1, 0, 1, 0, 1]
, fixing now.
You're welcomeXD
Got it. Thanks for your codes, too. They are really helpful.
So glad you find it useful :).
On Jun 29, 2017 09:58, "SongweiGe" notifications@github.com wrote:
Got it. Thanks for your codes, too. They are really helpful.
— You are receiving this because you modified the open/close state.
Reply to this email directly, view it on GitHub https://github.com/jma127/pyltr/pull/6#issuecomment-311974914, or mute the thread https://github.com/notifications/unsubscribe-auth/ACptEmpNJayihIVMNaVRKnw0tv8gJ2yAks5sI62hgaJpZM4OI6JV .
Hi Jerry,
I met AssertionError when running lambdamart with Z = MAP. Then I check the calculation on swap_deltas. And I found that there might be a mistake in your algorithm which caused the error. Maybe you can check it by use some examples.
Cheers