[Python] graph-tool crashes with segmentation fault at minimize_nested_blockmodel_dl() when...

I'm trying to run a weighted Stochastic Block Model (SBM) using graph-tool on a bipartite gene–SNP network, but I consistently hit a segmentation fault or ValueError: zero-size array at the line:

state = gt.minimize_nested_blockmodel_dl(g, state_args={"eweight": edge_weight})


Heres example code:


g = gt.Graph(directed=False)
vprop_type = g.new_vertex_property("int")
vmap = {}
eprop_weight = g.new_edge_property("float")

def add_vertex(name, ntype):
if name not in vmap:
v = g.add_vertex()
vmap[name] = v
vprop_type[v] = ntype
return vmap[name]

for _, row in df.iterrows():
v1 = add_vertex(row['left_node'], 0)
v2 = add_vertex(row['right_node'], 1)
e = g.add_edge(v1, v2)
eprop_weight[e] = row['weight']

g.vp["type"] = vprop_type
g.ep["weight"] = eprop_weight

state = gt.minimize_nested_blockmodel_dl(g, state_args={"eweight": eprop_weight})




I have:

• Created an undirected bipartite graph
• Assigned node types
• Assigned edge weights, all values in the range [0.1, 1.0], strictly positive
• Verified all edges and weights exist: number of edges > 27,000 No nodes are isolated
• Edge weight property was correctly added
• The unweighted SBM runs fine

Environment:

graph-tool version: (tried a couple...)

Python version: 3.13, 3.11, 3.10 (via Homebrew)

macOS: Ventura 13.x (Apple M1)

Graph has ~4,500 nodes, ~27,000 edges

Are there any known bugs related to edge-weighted SBM in graph-tool that I might be hitting?

Any tips for stabilizing or debugging weighted SBM inference on bipartite graphs?

Continue reading...