GraphWeb_DESI

GraphWeb_DESI builds a graph from DESI BGS galaxies and applies a pretrained GAT model (from TNG/Illustris) to infer cosmic web environments.

Main entrypoint

• Canonical: workflows/graph_inference/graph_catalog.py
• Compatibility wrapper: graph_catalog.py

This script now has an explicit CLI and no longer runs large ad-hoc plotting blocks by default.

Quick start

Run from repo root:

python graph_catalog.py

Canonical path form:

python workflows/graph_inference/graph_catalog.py

Other entrypoints follow the same pattern:

• workflows/catalog/load_catalog.py (shim: load_catalog.py)
• workflows/utilities/galaxy_catalog.py (shim: galaxy_catalog.py)
• workflows/utilities/investigate_edges.py (shim: investigate_edges.py)

Default behavior:

• graph type: alpha
• cache mode: rebuild
• writes VAC to GRAPHWEB_VAC_OUTPUT_PATH (from config)
• shows a summary histogram plot

Common run modes

Use cache if available, otherwise build:

python graph_catalog.py --cache-mode prefer-cache

Fail if cache files are missing:

python graph_catalog.py --cache-mode cache-only

Use Delaunay graph instead of alpha-complex:

python graph_catalog.py --graph-type delaunay

Run without summary plot:

python graph_catalog.py --no-summary-plot

CLI flags

Key options:

• --graph-type {alpha,delaunay}
• --cache-mode {rebuild,prefer-cache,cache-only}
• --first-moment-matching
• --cache-dir <path>
• --model-path <path>
• --scaler-path <path>
• --vac-output-path <path>
• --reference-catalog-path <path>
• --no-summary-plot

Path and environment configuration

Defaults come from config_paths.py. Useful env vars include:

• GRAPHWEB_CACHE_DIR
• GRAPHWEB_VAC_OUTPUT_PATH
• ILLUSTRIS_REPO_ROOT
• ILLUSTRIS_GAT_MODEL_PATH
• ILLUSTRIS_SCALER_PATH
• TNG_REFERENCE_CATALOG_PATH

Example:

export GRAPHWEB_CACHE_DIR=/pscratch/sd/d/dkololgi/graphweb_desi/cache
export GRAPHWEB_VAC_OUTPUT_PATH=/pscratch/sd/d/dkololgi/graphweb_desi/outputs/DESI_BGS_PRERELEASE_VAC.pkl
python graph_catalog.py --cache-mode prefer-cache

Outputs

Primary output:

• VAC pickle with predicted environment class and class probabilities.

Cached artifacts (in GRAPHWEB_CACHE_DIR):

• graph object (*_graph.pt)
• torch geometric data (*_geom.pt)
• scaled features (*_features.pt)
• zcat snapshot (*_zcat.pt)