How to use graynet results¶

graynet 2.0 stores results as run-level Parquet tables plus JSON metadata. GraphML and CSV are optional export formats, not the canonical on-disk layout.

Loading a run¶

from graynet import load_run

run = load_run("/path/to/run_dir")
print(run.metadata.keys())
print(run.raw_edges)

Filtering raw edges¶

from graynet import get_edge_values

edge_rows = get_edge_values(
    run.raw_edges,
    subject_id="sub-001",
    base_feature="freesurfer_thickness",
    weight_method="manhattan",
)

Converting to an ML matrix¶

X = run.raw_edges.to_ndarray(
    ["sub-001", "sub-002", "sub-003"],
    base_feature="freesurfer_thickness",
    weight_method="manhattan",
)

Reconstructing a NetworkX graph¶

from graynet import export_to_nx

graph = export_to_nx(edge_rows)

# centroid coordinates are added to node attributes when available
print(graph.nodes["lh_bankssts"])

Iterating subject by subject¶

for subject_id, subject_edges in run.raw_edges.iter_subjects():
    print(subject_id, subject_edges.to_pandas().shape)

Optional exports¶

If you need GraphML or CSV, use the CLI export subcommands documented in Command line interface.

Canonical tables are a better starting point for:

machine-learning matrices
subject-wise filtering
tabular analysis in Python or R
reproducible export to GraphML or CSV later