README.md

metadata

title: New Graph Visualization
emoji: 🕸️
colorFrom: blue
colorTo: green
sdk: docker
app_port: 7860
startup_duration_timeout: 1h
pinned: false

New Graph Visualization

Interactive clinical graph explorer (Dash + Cytoscape). FalkorDB and the Dash app run in one Docker container — same pattern as UMLS Graph Explorer.

On each cold start the container loads 1.28M nodes and 7.6M edges from staging parquet via GRAPH.BULK (~2–5 min), then serves the UI on port 7860.

Live Space: Raapid/new-graph-visualization

Default seed: A27828238 · Default hop depth: 2

Features

• Explore by AUI or CUI with hop depth 1–15 (HIER_ISA + MRREL)
• Toggle MRHIER, semantic (MRREL), AUI→CUI, same-CUI edges
• Path finder between two AUIs
• SAB / TUI / TTY / category filters (category from graph_nodes.parquet)
• Layout controls (cose-bilkent recommended)

Hugging Face Space (Docker)

Variable	Default	Description
PORT	7860	Dash listen port
FALKORDB_GRAPH	clinical_graph	Graph name
REBUILD_GRAPH	1	Reload graph on container start
SKIP_GRAPH_BUILD	0	Set 1 to skip load (debug)
STAGING_DIR	/home/user/app/falkor_staging	Staging parquet directory
REDIS_ARGS	--maxmemory 9gb …	Embedded redis-server memory

Set Space hardware to CPU basic (16 GB RAM) or higher.

One-time data upload (not in git — ~235 MB total)

SPACE=Raapid/new-graph-visualization
BASE="/media/anurag/Hard Disk/Graph New"

hf upload $SPACE \
  "$BASE/falkor_staging/concepts.parquet" falkor_staging/concepts.parquet \
  "$BASE/falkor_staging/mrrel_edges.parquet" falkor_staging/mrrel_edges.parquet \
  "$BASE/falkor_staging/hier_edges.parquet" falkor_staging/hier_edges.parquet \
  "$BASE/graph_nodes.parquet" graph_nodes.parquet \
  --repo-type space \
  --commit-message "Add graph staging parquets"

Local Docker

cd "/media/anurag/Hard Disk/Graph New"
docker build -t new-graph-viz .
docker run --rm --memory=16g -p 7860:7860 new-graph-viz

Open http://localhost:7860.

Requires falkor_staging/*.parquet and graph_nodes.parquet in the build context (included if present; bulk_csv/ is excluded via .dockerignore).

Local dev (FalkorDB in separate container)

docker run -d --name clinical-falkordb -p 6380:6379 --memory 10g falkordb/falkordb:latest
export FALKORDB_PORT=6380 FALKORDB_GRAPH=clinical_graph
python3 build_falkor_graph.py --skip-prepare --rebuild --falkor-port 6380
export PORT=8050
python3 app.py

Data

File	Role
falkor_staging/concepts.parquet	FalkorDB Concept nodes
falkor_staging/mrrel_edges.parquet	MRREL semantic edges
falkor_staging/hier_edges.parquet	HIER_ISA (from MRHIER)
graph_nodes.parquet	Category metadata for sidebar filter

FalkorDB node count (1,278,278) matches edge-connected nodes in parquet (orphan nodes excluded at load time).

Project layout

Graph New/
├── app.py
├── graph_backend.py
├── Dockerfile
├── scripts/entrypoint.sh
├── falkor_load/           # GRAPH.BULK loader (container startup)
├── falkor_staging/        # staging parquet (upload to HF separately)
├── prepare_falkor_staging.py
├── build_falkor_graph.py  # local load helper
├── graph_nodes.parquet
├── assets/
└── requirements.txt

Commands

# Setup
uv venv .venv && source .venv/bin/activate
uv pip install -r requirements.txt

# Rebuild staging from raw parquet
python3 prepare_falkor_staging.py --connected-only

# Load FalkorDB locally
FALKORDB_BULK_SOCKET_TIMEOUT_S=0 python3 build_falkor_graph.py --skip-prepare --rebuild --falkor-port 6380

# Run app
FALKORDB_PORT=6380 FALKORDB_GRAPH=clinical_graph python3 app.py