attributes error- graph_nets_basics issue

[ghost]

When running the graph_nets_basics.ipynb on Colab, the following code got AttributeError. Thanks.

tf.reset_default_graph()
graph_network = modules.GraphNetwork(
    edge_model_fn=lambda: snt.Linear(output_size=10),
    node_model_fn=lambda: snt.Linear(output_size=15),
    global_model_fn=lambda: snt.Linear(output_size=20))

input_graphs = utils_tf.data_dicts_to_graphs_tuple(graph_dicts)
output_graphs = graph_network(input_graphs)

for var in graph_network.variables:
  print(var)

AttributeErrorTraceback (most recent call last)

in () 8 output_graphs = graph_network(input_graphs) 9 ---> 10 for var in graph_network.variables: 11 print(var) AttributeError: 'GraphNetwork' object has no attribute 'variables'

[alvarosg]

Thank you for pointing this out.

I think this issue may be related to the fact that the library is currently pinned to sonnet v1.23 (see issue).

This is in the process of being fixed, but you may try upgrading sonnet manually in the meantime: pip install dm-sonnet --upgrade (or !pip install dm-sonnet --upgrade if you are in Colaboratory).

Alternatively you may use graph_network.get_variables().

Hope this helps!

[ghost]

It works! Thanks!