Dynamic SPN example

[anicolson]

Hi Jos,

I notice that you have started implementing dynamic SPNs: https://github.com/pronobis/libspn-eras/blob/master/libspn_keras/layers/temporal_dense_product.py

Thats great!

I was wondering if you could give a simple tutorial about how to implement a dynamic SPN.

Thanks.

[jostosh]

Hi Aaron, I'm definitely planning to do this. Any suggestions for a dataset?

[jostosh]

Thanks for the suggestion. I'm afraid your example does not really resemble the typical maximum likelihood training scenario for which SPNs lend themselves well. Probably, one could still do something to similar to what you propose with SPNs, but it is to me not yet trivial how I should proceed.

I'm working on getting dynamic SPNs to the forefront as we speak, but in the process, I stumbled on some false assumptions I made about the things I built so far and the compatibility with TensorFlow 2.x so it's taking a little bit longer than expected, but it should be there later ths week. I will include one of the datasets that's also used in the dynamic SPN paper.