How to get and set the weights of neural networks in GNN about graph_nets HOT 5 CLOSED

After you connect the module for the first time:
output_graphs = graph_net_module(input_graphs)

Then you can use graph_net_module.get_variables() to recover all of the variables containing the weights, including those of the edge model.

from graph_nets.

Closing due to inactivity.

from graph_nets.

After you connect the module for the first time:
output_graphs = graph_net_module(input_graphs)

Then you can use graph_net_module.get_variables() to recover all of the variables containing the weights, including those of the edge model.

Hey,
thanks for your cool library!
I have a question regarding the graph_net weights. I've copy-pasted the example posted above and presented an own graphTuple as input. When trying to extract the weights as described above, the following error occurs:

AttributeError: 'GraphNetwork' object has no attribute 'get_variables'

How do I get the weights?

from graph_nets.

Hi, are you using the tensorflow 2 version? is so you can access the variables using: graph_net_module.trainable_variables after you connect for the first time.

You can access all variables using graph_net_module.variables.

from graph_nets.

Hi, are you using the tensorflow 2 version? is so you can access the variables using: graph_net_module.trainable_variables after you connect for the first time.

You can access all variables using graph_net_module.variables.

Yes, I am using tensorflow 2 and your proposed solution solved my issue. Thank you for the fast and helpful reply!

from graph_nets.