Sorting dataset based on the nodes' page rank values

[OliverSummers]

I asked a similar question in another post, and according to the provided answer to sort the Planetoid dataset based on nodes' PageRank, I coded as follows:

 import networkx as nx
 from torch_geometric.utils.convert import to_networkx
 from torch_geometric.datasets import Planetoid
 
 dataset = Planetoid('./data', 'PubMed')
 data = dataset[0]

 nx_data = to_networkx(data)
 ranks = nx.pagerank(nx_data, alpha=0.9, tol=1.e-08)
 pagerank = list(ranks.values())
 pagerank_tensor = torch.tensor(pagerank)
 perm = pagerank_tensor.argsort()

 inv_perm = torch.empty_like(perm)
 inv_perm[perm] = torch.arange(perm.size(0))

 new_x = data.x[perm]
 new_y = data.y[perm]
 new_edge_index = inv_perm[data.edge_index]

 data.x = new_x
 data.y = new_y
 data.edge_index = new_edge_index

 dataset.transform = data

Then I check the transformed dataset to see if it is sorted based on the nodes' PageRank. The following snippet prints out the indices of the consecutive nodes with the same PageRank values:

 trans_data = dataset[0]
 nx_graph = to_networkx(trans_data)
 ranks = nx.pagerank(nx_graph, alpha=0.9, tol=1.e-08)
 rank_list = list(ranks.values())

 s = set(rank_list)
 s_list = list(s)
 dict_inx = {}

 for i in s_list:
    first_index = ran_list.index(i)

    reversed_list = ran_list[::-1]

    first_index_in_reversed = reversed_list.index(i)

    last_index = len(ran_list) - 1 - first_index_in_reversed

    dict_inx[i] = (first_index, last_index)
 
 for k, v in dict_inx.items():
   print(k, v)

The output of the above snippet is:

  1.5705199615468062e-05 : (3339, 3340)

  5.347380017895053e-05 : (15102, 15102)

  2.597480481370294e-05 : (10631, 10634)

  .

  .

  .

  2.672946705217494e-05 : (10935, 10935)

  2.771347590169145e-05 : (11333, 11333)

where the number before : is PageRank and the tuple after : denotes indices of all nodes with the same PageRank.
As we can see, the page ranks are sorted in ascending order but their corresponding indices are not sorted. If all of the above code is written based on the nodes' degree values then the final output will be:

 1 : (0, 9093)

 2 : (9094, 12450)

 3 : (12451, 14034)

 .

 .

 .

 154 : (19715, 19715)

 171 : (19716, 19716)`

where the number before : is degree and the tuple after : refers to the nodes' indices with the same degree values.
So as we can see, the results with the sorted dataset based on degree values are perfect. Both degrees and indices are sorted ascendingly but as I explained above, it is not the case with PageRank.
My question is how to fix the problem of the PageRank code?

[rusty1s]

Sorting of the inner lists is not really necessary, but if you want to do that I suggest iterating over each page rank value and sorting the list of node indices manually, e.g.,:

for k, v in dict_inx.items():
   dict_inx[k] = sorted(v)

[OliverSummers]

v tuple is already sorted but my point was how to sort the tuples one after the other. For example in sorting based on degree, the nodes with degree 1 correspond to index 0 to 9093, and then with degree 2 it starts at 9094 to 12450 however in PageRank case, in the transformed dataset, the smallest page rank does not start at index 0 but at index 3339 and then the next page rank starts at index 15102. I hope I could clarify my question better.

[rusty1s]

I don't see a problem here. Simply converting/flatten your dictionary into a list sorted based on PageRank value
will give you a similar vector as what you would have received after calling degree.argmax(), i.e., the permutation of the nodes so that they are sorted by PageRank value.