Calculation of the edge_attr

[domalbert]

Using a custom image dataset I'm trying to train a graph neural network by first transferring it via SLIC segmentation using ToSLIC and approximating the edge index using KNNGraph. Is there a possibility to determine the edge_attr.

My dataset looks something like this:

    def __init__(self, root, transform=None, pre_transform=None, pre_filter=None):
        super().__init__(root, transform, pre_transform, pre_filter)         
        x = np.random.randint(0,255, (100,25,25))
        y = np.random.randint(0,3,(100,))
        
        self.x = torch.Tensor(x)
        self.y = torch.Tensor(y)
        
        #conversion in slic
        self.data = Data()
        for i in range(len(self.x)):
            slic = ToSLIC()(self.x[i].reshape(1,25,25))
            knn = KNNGraph()(slic)
            
            self.data.x = knn.x
            self.data.edge_index = knn.edge_index
            self.data.pos = slic.pos
            #self.data.y = y
            self.data.num_classes = int(torch.max(self.y))
            
        
        
    def process(self):
        return self.data, self.y
        
    def _download(self):
        return

    def _process(self):
        return

    def __repr__(self):
        return '{}()'.format(self.__class__.__name__)


my_dataset = MyDataset('/data') # create your datset
my_dataloader = DataLoader(my_dataset, batch_size=8) # create your dataloader

Can any one give me guidance on how I'm able to determine the edge_attribute?

[rusty1s]

You can generate edge features via additional transforms, e.g., transforms.Cartesian. This will give you the relative distances between nodes as edge features.

[domalbert]

Thank you. This solved my problem.