Reset Labels for Labeled Nodes in LPA

[unbreakablegao]

Hi,

In "model.py" when building lpa as following code, the labeled nodes seem not to be reset in the loop. I am wondering if this is correct. thank you! def _build_lpa(self): label_mask = tf.expand_dims(self.label_mask, -1) input_labels = label_mask * self.labels label_list = [input_labels]

    for _ in range(self.args.lpa_iter):
        lp_layer = LPALayer(adj=self.normalized_adj)
        hidden = lp_layer(label_list[-1])
        label_list.append(hidden)
    self.predicted_label = label_list[-1]

[jlidw]

@unbreakablegao @hwwang55 I am reading this code, and have the same question...

[bozhenhhu]

me too

[Hydrapse]

I think the key point is, the input label of LPA (e.g., label[train_mask]) is the same as the ground truth label, which seems analog to an auto-encoder.