MatrixGraph is a C++/CUDA library for parallel graph computing. This directory contains documentation for tools, GPU tasks, and CPU tasks.
| Document | Description |
|---|---|
| GraphConverter | Format conversion (CSV, edgelist, CSR, tiled matrix, EGSM, VF3, etc.) |
| GraphPartitioner | Graph partitioning (GridCut) for tiled processing |
| FormatConverter | Internal C++ format conversion utilities |
| Preprocessing4MatrixFilter | Python preprocessing for ML filter (Rapids→torch, embeddings, training) |
| GenerateRandomGraph | Random vertex-labeled graph generator |
| SubIsoTraining | ML filter training workflow for SubIso |
| Embedding | Graph embedding generator (PyTorch → binary) |
| ComputeF1 | F1 / Precision / Recall calculator |
| Document | Description |
|---|---|
| SubIso | Subgraph isomorphism (VF3 + ML filter) |
| Document | Description |
|---|---|
| Matrix Operations | Matmult, Activate (ReLU) |
Graph → WCC / BFS / PageRank:
CSV → graph_converter (edgelistcsv2edgelistbin) → edgelist
edgelist → graph_converter (edgelistbin2csrbin) → CSR → wcc_exec / bfs_exec / pagerank_exec
Graph → GEMM / PPR (tiled):
CSV → graph_converter (edgelistcsv2edgelistbin) → edgelist
edgelist → graph_partitioner (gridcut) → partitions
partitions → graph_converter (gridedgelistbin2csrtiledmatrix) → tiled → gemm_exec / ppr_query_exec
SubIso ML filter training:
Text graph → graph_reader.py → .pt → data.py → embedding
Ground truth (custom) + embeddings → train.py → model