Making the data available?

[olagustafsson1]

I read the great paper and was eager to try out the model. Having the data available would be helpful. Any plans on posting the datasets or references to where they can be found?

[liuchbryan]

Sorry for the late reply.

Unfortunately we are still not allowed to post the dataset involved, as you probably understand that contains some sensitive information of users from ASOS.

Georg made the following comment a while ago though when answering a similar query:

I generate the dataset for training and testing here: https://github.com/grobgl/rnnsm/blob/master/code/rmtpp/rmtpp_data.py.
This results in a 3-dimensional input array x representing [users * sessions * session_features]. I use a pre-denfined maximum number of sessions per user (default 64) and pad users with fewer sessions than that with zeroes. Note that categorical features (device, hour of day, day of week, day of month) are encoded as integers 1..n.

The target values are a 2-dimensional array y where each row contains:

• a value representing the time interval between a user's latest session in the observation window and their first session in the prediction window (or the end of the prediction window if they do not return)
• a value set to 0 if the user returns in the prediction window and 1 if the user does not return (the observation is censored)

Hope that helps to get you started!