Implementation of Multitask Multiple-Kernel Learning in python using shogun

Question

I'm a chemistry student new to the community interested in implementing the paper of Widmer & Raetsch 2012. Particularly, I would like to implement task similarity learning by using powerset approach. I have around ~1500 examples classified in 10 tasks, of which I would like to compute pairwise task similarities. It looks like

1. setting custom kernels to mask only certain subtasks
2. for a certain pair of examples, to add all subtask kernels that the examples are part of
   are what's important, but I have no idea how to do this using Shogun in python. Can anybody please guide me to an example code or a tutorial that I can look into?

Thank you so much in advance!