About prior structure

[Beck-Pro]

First of all, thanks for this amazing paper and code.
so the picture above is the prior structure I'm trying to implement.
(competition is from this : https://dacon.io/competitions/official/236423/data)

• about data
  time stamp : per minute (could be irregular)
  Q : flux (Q1-Q5)
  M : Pump wheter On=1 or Off=0 (M1-M14) (these are located in Pumping station in the 2nd rightmost side of picture)
  P : Pressure (P1-P26)
  anomaly : normal=0 or abnormal=1
  P_flag : Whether Pressure at that time is normal=0 or abnormal=1
  (P_flag_1-P_flag_26)

In 'def get_prior_graph_struc(dataset)' function,
it seems i have to customize the structure of graph based on the picture.
any advice doing that?

[Beck-Pro]

Oh and plus, so train data is Both TRAIN_A and TRAIN_B that have both different graph structure. not only that, have 2 test sets TEST_C and TEST_D that have both different graph structures too.

so same model have to learn 2 different graph structures with 2 different train sets. any advice about training method?