Gorot method

[larissayukimiz]

Hi!
I'd like to know why you use output_dims as the fan-in in f's and g's (at the Gorot method). I mean, why not use, as the fan-in, hid_dims and as the fan-out, the output_dims?

[hongzimao]

Good question. Notice that the GNN structure (loosely) is h -> f -> g. So the dimensions in the connections are input_dim -> ouput_dim (kind of like a "hidden_dim") -> output_dim -> output_dim. The "hidden_dim" should not be confused with self.hid_dims, which is within each h, f, g. You can track how h, f, g are actually used by searching when prep proc and agg are called. Hope this helps!