Xiaoyao Yang

@casperdcl Thanks for the reply, I don't think so though.. I read the docs it says > Flag for outputting the notebook without execution, but with parameters applied. Can you...

> ranking_model I think what you suggest would work and it is similar to what I did (by random gues..), I would like to know more **_what exactly the model...

Positive interaction would click, negative interaction would be impressions without clicks, neural interaction would be no clicks and no impressions. https://github.com/maciejkula/triplet_recommendations_keras In terms of using TFRS, I am not sure...I...

@ydennisy yeah.. #334 has relevant discussion: in your example, the matrix would look at this. | | stereo | hifi | hifi | cd plaer | |-------|--------|------|------|----------| | bob |...

> This is called an in-batch softmax loss. The input data is as follows: > > ``` > # User embeddings # Candidate embeddigns > [[user_0], [[candidate_clicked_by_user_0], > [user_N]] [[candidate_clicked_by_user_N]]...

> Hi @xiaoyaoyang, It's still an identity matrix. > There may be candidates that the user has interacted with which are used as negatives for this batch, even though they...

> ``` > (u1,i3) (u1, i3) > (u2,i3) (u2, i3) > ``` > > Here, the negative for each positive case is the same as the positive item. This is...

Did you get any warning or error? I got some warning saying it will reduce it to 1 GPU/CPU.

@kim-sardine thanks! the tfrs.Model use tf.keras.Model though.... will give it a try Seems it is a known issue, replacing with `tf.distribute.MultiWorkerMirroredStrategy` seems to be a walkaround.

@italodamato Accuracy is the `metric` and `val_loss` is the loss. The algo only optimize the loss, metric is whatever you define.