Why do geographical embedding and semantical embedding share the GCN parameters in GEML?

[Ao-Last]

According to the paper, geographical embedding and semantical embedding are trained through two sperate networks. But ge and se are generated using the same GCN module in your implementation libcity/model/traffic_od_prediction

[aptx1231]

@l782993610

[Apolsus]

You are right, thank you. It's fixed in pr #323