@jmgilmer Trying to reach out

[hochshi]

@jmgilmer Hi,

I'm trying to figure out why this line appears in the code.

mpnn/set2vec.py

Line 156 in 4a1f0dd

attention_w2 = tf.get_variable(name + "att_W_2", [node_dim, node_dim])

It doesn't appear in the MPNN or vinyalis et al. (2016) Order Matters, as far as I can understand. Can you please correct me or provide clarification as to why include this added attention matrix?

Thank you,
Shi