discopt.decomposition.graph.export

discopt.decomposition.graph.export#

Serialize model-induced graphs to interchange formats.

The advisor’s export_graph surface (design doc §14) dumps a graph for external tooling — graph databases, partitioners, and visualizers. Four formats cover the immediate needs:

  • "json" — a node-link dict, easy to consume anywhere.

  • "dot" — Graphviz, for quick rendering.

  • "graphml"— for graph-analysis tools (Gephi, networkx, igraph).

  • "metis" — the graph input format for the METIS/KaHyPar partitioners; feed it to an external partitioner to obtain a bordered-block-diagonal ordering, then re-import the result via the GCG-style .dec reader (read_dec()).

Exports operate on the variable dependency graph (variable vertices, edges = co-occurrence in a constraint), which is what partitioners consume. The bipartite incidence is available via the bipartite flag for visualization.

.dec interchange (write_dec() / read_dec()) round-trips a resolved decomposition to/from the format SCIP and GCG use, so an external partitioner or GCG run can supply the block/border structure.

Functions#

export_graph(→ str)

Serialize graph's variable dependency graph to fmt.

write_dec(→ None)

Write a GCG-style .dec file for structure over model.

read_dec(...)

Read a GCG .dec file and return a DecompositionStructure.

Module Contents#

discopt.decomposition.graph.export.export_graph(graph: discopt.decomposition.graph.base.ModelGraph, fmt: str = 'json') → str#

Serialize graph’s variable dependency graph to fmt.

Parameters:

  • graph (ModelGraph) – The model-induced graph to export.

  • fmt (str) – One of "json", "dot", "graphml", "metis".

Returns:

The serialized graph.

Return type:

str

discopt.decomposition.graph.export.write_dec(structure, model, path: str) → None#

Write a GCG-style .dec file for structure over model.

Emits NBLOCKS, one BLOCK k section listing that block’s constraint identifiers, and a MASTERCONSS section for the coupling rows — the format SCIP/GCG read and write, so a discopt decomposition can be handed to (or taken from) that toolchain.

discopt.decomposition.graph.export.read_dec(path: str, model) → discopt.decomposition.structure.DecompositionStructure#

Read a GCG .dec file and return a DecompositionStructure.

The MASTERCONSS become the coupling rows; blocks are recomputed from the non-coupling constraints (so the partition is always internally consistent with the model, regardless of how the file grouped them).