Explanation of loss function

[rragundez]

Doesn't the WMRBLossGraph return a tensor and not a real number? How can the optimzer work with that? I mean the line 177 sampled_margin_rank = (n_items / n_sampled_items) * tf.reduce_sum(summation_term, axis=1) is reducing in the columns dimension, so you will still have a 1D tensor there.

and is the tf_sample_predictions computed for each forward pass?

https://github.com/jfkirk/tensorrec/blob/220c51af656b8930abebf01641e2a936952c45c6/tensorrec/loss_graphs.py#L137