Wrong dimension in sparse incidence matrix about xgi

[leotrs]

The problem is this line: https://github.com/ComplexGroupInteractions/xgi/blob/388114d72d3ba03ea7c04ced52532762f3588ef8/xgi/linalg/matrix.py#L76 The sparse matrix is being created only from edges that have members (and therefore only from edges that have memberships).

…

On Thu, Apr 7, 2022, 10:51 AM Maxime Lucas ***@***.***> wrote: The incidence matrix should always have dimension (num_nodes, num_edges) or (num_nodes, num_edges_of_order). But when sparse=True, dimension 0 is wrong when some nodes are disconnected (overall, or at a given order when it is specified). This makes all the other functions using the incidence matrix wrong (adjacency, laplacian, etc.). MWE: nodes = range(5)edges = [[0,1], [0,2], [1,2,3]] H = xgi.empty_hypergraph()H.add_nodes_from(nodes)H.add_edges_from(edges) xgi.incidence_matrix(H, sparse=False).shape## --> (5, 3) # goodxgi.incidence_matrix(H, sparse=True).shape## --> (4, 3) # not good This is because when sparse=True, the sparse matrix is built by iterating over the members of edges (of order d): https://github.com/ComplexGroupInteractions/xgi/blob/388114d72d3ba03ea7c04ced52532762f3588ef8/xgi/linalg/matrix.py#L71-L82 — Reply to this email directly, view it on GitHub <#72>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAILYAFRONXMFDSYYN4AYODVD2OY3ANCNFSM5SYT4ONA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[maximelucas]

Yes exactly.

If we want to use the same csr_matrix constructor csr_matrix((data, (rows, cols))) , then we need to iterate over all possible edges (including that do not exist). This doesn't seem optimal.

Another solution would be to build the non-sparse matrix (which already do in the sparse=False case), and then say if sparse: I = csr_matrix(I) using another constructor. Not sure how (in)efficient it would be to convert from dense to sparse this way.

Any other idea? The list the constructors here: https://docs.scipy.org/doc/scipy/reference/generated/scipy.sparse.csr_matrix.html

[leotrs]

You could iterate over all the nodes (even the ones that are isolated) and then iterate over the edges of each node. No need to iterate over inexistent nodes or edges.

[maximelucas]

Good idea!