as the for bpr loss function, is the expression right in the code?

[jackyhawk]

As for the loss function and updating the latent vector(parameter) ,

i find it is different implementment in the code and in the pater

in code:
Double BPREngine::loss(const Double scoreDifference) const {
return log(1.0 + exp(-scoreDifference));
}

in paper:

i just paste the rule here:
http://photo27.hexun.com/p/2019/0201/632421/b_vip_11118BF2E45A1C8033DC6DA3576F7A2C.jpg
http://photo27.hexun.com/p/2019/0201/632421/b_vip_11118BF2E45A1C8033DC6DA3576F7A2C.jpg

you can get the rule here

Θ = Θ + α( e^(-Xuij) / (1+ e ^(-Xuij) )........)

[albietz]

Hi, the function loss just gives you the loss for a given triplet, which in this case is given by the negative log-likelihood according to the BPR model. What you're showing in the image corresponds to gradient updates, which you can find in the BPREngine::update function (see https://github.com/quora/qmf/blob/master/qmf/bpr/BPREngine.cpp#L178).

Hope this helps,
Alberto

[jackyhawk]

Thanks very much Alberto.

And I just made a mistake, what i want to say is the lossderitive in the training presss,which is
https://github.com/quora/qmf/blob/master/qmf/bpr/BPREngine.cpp#L243
https://github.com/quora/qmf/blob/master/qmf/bpr/BPREngine.cpp#L183

just as you mentioned in the update function.

however the lossDerivative is not the same as the highlight part in the picture:

http://photo27.hexun.com/p/2019/0201/632421/b_vip_11118BF2E45A1C8033DC6DA3576F7A2C.jpg

i wonder whether we should use this expression:

1.0 / (1.0 + exp(scoreDifference))

or this expression:
exp(-scoreDifference) / (1.0 + exp(-scoreDifference)); (and it is the same as: e^(-Xuij) / (1+ e ^(-Xuij) ) )

to calculate the loss deritive?

[albietz]

Those two expressions are in fact equal :)

[jackyhawk]

Got it, Thanks very much Alberto.