Michaël Defferrard

I'm facing that issue as well. Probably related, applications would set an urgent flag when requesting focus, and I could jump to them with `bindsym $mod+x [urgent=latest] focus`.

That's probably due to a TensorFlow update. The original code should work with `tensorflow-gpu==1.1.0` ([the version I used to develop the code](https://github.com/mdeff/cnn_graph/commit/f7f7ba1e0094642fce4353dbe53d8809c98ada22)). If the fix (adding `tf.app.flags.DEFINE_string('f', '', 'kernel')`) also...

It was indeed developed with signal classification in mind. For node classification, you mainly have to change the objective function. The adjacency matrix A only depends on the graph, whatever...

Thanks. It should be `assert len(indices[0]) == M` right?

Me too 🙈 For reference, here's what I have from a later iteration of this code: ```python def test_permutations(): parents = [np.array([4, 1, 1, 2, 2, 3, 0, 0, 3]),...

Thanks! Does that work with `gensim==2.1.0` ([the version used during the development of this code](https://github.com/mdeff/cnn_graph/commit/f7f7ba1e0094642fce4353dbe53d8809c98ada22))?

It will almost work. ;) You'll however need to make some small changes. The first is to comment the line `x = tf.expand_dims(x, 2) # N x M x F=1`...

Hi Maosi, Regarding definitions, you're correct that the diagonal of the combinatorial Laplacian `L = D - W` is zero on the fake nodes. It is however undefined for the...

Then it does not matter indeed. The method was clearly developed with a fixed graph in mind, but filters can definitely be applied to multiple graphs (especially if they are...