Best way to associated node features in graph with node features loaded from disk

[DanielPerezJensen]

I have the following definition of a heterogeneous graph:

import torch
import dgl

subsub_src = [10, 8, 5, 7, 6, 0, 1, 2, 3]
subsub_dst = [11, 9, 9, 9, 9, 1, 4, 3, 4]
msrmsr_src = [1, 2, 3]
msrmsr_dst = [0, 0, 0]
submsr_src = [4, 9, 11]
submsr_dst = [1, 2, 3]

data_dict = {
    ('subsubwatershed', 'flows', 'subsubwatershed'): (subsub_src, subsub_dst),
    ('measurement', 'flows', 'measurement'): (msrmsr_src, msrmsr_dst),
    ('subsubwatershed', 'in', 'measurement'): (submsr_src, submsr_dst)
}

g = dgl.heterograph(data_dict)
print(g)

Each different type of node and edge has a different type of feature set that I wish to add to the graph as well. What would be the easiest way to load that in. For example in the measurement nodes I wish to add the following information:

      Centroid_X  Centroid_Y  altitude
node                                  
0        123      321      11
1        234      432      22
2        345      543      33
3        456      654      44

In the dgl heterogeneous graph user guide it is stated that you can simply add features using g.nodes['measurement'].data['hv'] = th.ones(3, 1) for example.

This data is of course randomly generated, but I don’t see an easy of constructing this from data loaded from disk as in my usecase.

It seems like I almost will have to hardcode every feature to be in the proper order based on my graph, how do I go about this easily? Is there some sort of built-in function to achieve this?

EDIT:

I wrote the following function to do it.

import copy

def add_node_features(graph: dgl.DGLGraph, node_type: str, features: pd.DataFrame):
    new_graph = copy.deepcopy(graph)
    
    for feature in features:
        new_graph.nodes[node_type].data[feature] = (torch.from_numpy(features[feature].to_numpy()))
    
    return new_graph

static_measurement_nodes_df = pd.DataFrame(
                                    {
                                        "Centroid_X": [123, 234, 345, 456], 
                                         "Centroid_Y": [432, 432, 543, 654],
                                         "altitude":   [11, 22, 33, 44]
                                    }
                                )

g = add_node_features(g, "measurement", static_measurement_nodes_df)
print(g.nodes["measurement"])

But this of course assumes that the order of the given dataframe is in the same order as in whatever order the .nodes attribute returns. What if these orders are not the same?

And how do I go about checking that indeed the correct nodes have the correct associated data?

[BarclayII]

DGL assumes that the node IDs are always ordered as consecutive integers starting from 0 (i.e. 0, 1, …). If your dataframe is not ordered like that, is it possible for you to reorder the dataframe?