Train test split

[Jarvisss]

Thanks for release the code!
I notice that the current code in create_tuples_dataset.py makes a random split.
So could you please share your train-test split file same with the paper?

[Jarvisss]

Looks like most of ids in fashion dataset have 4 or less images available.
I'm curious why 6 input images is used for fashion.

[Jarvisss]

@Stephlat
Set nb_inputs=6 for fashion dataset returns null list.