How to pass sequential events with times to DSIN?

[klimentij]

Thank you for this amazing library! I have a training set where each row is a series of user actions on a website (logged in, sent an invoice, etc.) with times deltas in ms between these actions. Each row has a label — a corresponding user class (10 classes):

action132, 2340, action21, 300, ...  3
action238, 1240, action22, 350, ...  6
...
action763, 1240, action42, 750, ...  2 

Seems to be the perfect case for DSIN. As I understand after reading the DSIN paper, I have to pass the sequence of all user actions along with times (converted to times since the first user's timestamp) to Session Divsion Layer and this layer supposed to split by itself this long sequence to sessions (split on >30 min gaps). But would be the correct format for the input vector? Should it be 3D with times under actions, should I leave time between actions or just flatten it out so that all events go first, then go all times?

The other confusing thing is that as I understand I have to pass it as VarLenSparseFeat. But the problem is times are Varlen, but not sparce and I didn't see "VarLenDenseFeat" class in your API.

It would be awesome if you give a quick example of how to pass such data correctly to DSIN. Even better if you add it to your DSIN code example so that everybody can benefit from it.

Thank you!

[Lulu20220]

I have the same question, how to pass Time features? Have it been solved?