About dataset format and running the code

[Abhinav43]

Hello Ma,

I am trying to run your code but it requires

• 'allx',
• 'ally',
• 'graph',
• "adjmat",
• "trainMask",
• "valMask",
• "testMask"

I checked the preprocessing files and https://github.com/matenure/FastGCN/tree/master/data
for data format but couldn't find enough resources.

Since most of the graph convolutional networks are based on node prediction so their format are different i think?

Can you provide a simple preprocessing script with few artificial datapoints, That would help to understand the shape of all format placeholders.
If I am trying to run this code on cora dataset I am getting error :

ValueError: shapes (1,2708) and (1,1) not aligned: 2708 (dim 1) != 1 (dim 0)

my data shapes look like this:

• print(np.array(adjs).shape), >> (2709,)
• print(features.shape) , >> (1708, 1433)
• print(x.shape) , >> (1708, 1433)
• print(y.shape) , >> (1708, 7)
• print(train_mask.shape) , >> (2708,)
• print(test_mask.shape), >> (2708,)
• print(val_mask.shape) >> (2708,)

Can you share the correct shape format?

Second I was going through the paper (https://arxiv.org/pdf/1804.10850.pdf) , Paper says :

Assume we have adjacency matrix Au for view u, we assign attention weights g u ∈ R N∗N to the graph edges, such that the integrated adjacency matrix becomes sigma u g u  A u where is the element wise multiplication.

But in implementation you are not using element wise multiplication, also I am not clear about how you are concatenating with 0? If i am getting right then final mixedADJ will be same shape as original adj?

def attention(self):
        self.attweights = tf.get_variable("attWeights",[self.num_support, self.output_dim],initializer=tf.contrib.layers.xavier_initializer())
        
        #self.attbiases = tf.get_variable("attBiases",[self.num_support, self.output_dim],initializer=tf.contrib.layers.xavier_initializer())
        attention = []
        self.attADJ = []
        for i in range(self.num_support):
            #tmpattention = tf.matmul(tf.reshape(self.attweights[i],[1,-1]), self.adjs[i])+tf.reshape(self.attbiases[i],[1,-1])
            tmpattention = tf.matmul(tf.reshape(self.attweights[i], [1, -1]), self.adjs[i])
            #tmpattention = tf.reshape(self.attweights[i],[1,-1]) #test the performance of non-attentive vector weights
            attention.append(tmpattention)
        print("attention_sie",attention.size)
        attentions = tf.concat(0, attention)
        self.attention = tf.nn.softmax(attentions,0)
        for i in range(self.num_support):
            self.attADJ.append(tf.matmul(tf.diag(self.attention[i]),self.adjs[i]))

        self.mixedADJ = tf.add_n(self.attADJ)

Thank you
Keep writing and keep sharing good work.
Looking forward to your reply, Thank you :)

[matenure]

Assume we have 222 drugs, here are the shapes for each variable:
adjs: a list with 4 elements, each adjs[i] with shape: (222, 222)
features/x: (222, 582)
y: (222, 222)
train_mask: (222, 222)
val_mask: (222, 222)
test_mask: (222, 222)

Yes, the final mixedADJ will have the same shape as original adj.

I can upload the masks and adjs in this repo for reference.