Steven Morad

We're looking at software options for SWIMSat: https://digitalcommons.usu.edu/cgi/viewcontent.cgi?referer=https://www.google.com/&httpsredir=1&article=3442&context=smallsat

Using batch mode, would this be ``` batch = Batch(x=torch.rand(num_graphs, n, num_feats), adj=torch.randint(0, 1, (num_graphs, n, n))) edge_index = (batch.adj > 0).nonzero().t() b, row, col = edge_index batch.edge_weight = batch.adj[b,...

Apparently, not all models take `edge_weight` as input, e.g. `GATConv`. Is there a way to propagate gradients in this case? I suppose one option would be extending said models to...

@vctorwei you may want to look at https://github.com/rusty1s/pytorch_geometric/issues/2543 I'm planning on upstreaming this after a paper deadline in ~3 weeks.

It would be nice to have an option to utilize the `gibson` "hole-filler" network in the rendering pipeline: https://github.com/StanfordVL/GibsonEnv/blob/8e9e23c2dcc44bb5442a25d12928103f05b51d8f/gibson/core/render/pcrender.py#L169

@dthkao is probably correct. Ensure that your switch is not a router running a DHCP server for the clients.

Unfortunately, not at this time.

We would. Please keep in mind we are no longer actively developing the `master` branch. If you want to add the feature, I would suggest adding it to the `golang`...

@mattjj thanks for the tip, I was unaware of the array caching mechanics (hidden side effect!). I figured precomputing all the keys in one batched operation would be more efficient...

Sorry I'm not super familiar with your PR process. https://github.com/google/jax/pull/17452 does not have any commits attached to it. I just cloned and tested on master and I'm still seeing the...