Skip to content

Data-format.md

Latest commit

 

History

History
94 lines (70 loc) · 2.33 KB

Data-format.md

File metadata and controls

94 lines (70 loc) · 2.33 KB
Raw

Data Format

EpiNet expects two CSV files: one for nodes and one for edges. The schema is domain-neutral so the same toolkit can be used for epidemiology, clinical pathways, research collaboration networks, referral systems, grant maps, patient journeys, or other graph-shaped problems.

Nodes

Required:

ID

Optional but commonly useful:

Outcome,Type,Label,Group,Time,Feature1,Feature2,...

Example:

ID,Outcome,Type,Age,Clinic
Patient_1,1,Patient,62,A
Patient_2,0,Patient,44,B
Clinic_A,,Clinic,,

Rules:

  • ID must be unique.
  • Outcome is optional. If present, it can be used for outcome modeling and for defining target nodes in shortest-path analysis.
  • Numeric node attributes are added to the optional outcome model.
  • Missing values should be meaningful. The toolkit does not assume that blank, unknown, not measured, and zero are the same thing.

Edges

Required:

SourceID,TargetID

Optional:

Weight,Relationship,Time,Direction,Metadata...

Example:

SourceID,TargetID,Weight,Relationship
Patient_1,Clinic_A,1.0,visit
Patient_2,Clinic_A,1.0,visit

Rules:

  • Every SourceID and TargetID must exist in the node file.
  • By default, edges are treated as undirected.
  • Pass --directed if SourceID -> TargetID direction matters.
  • Pass --weight-column Weight to copy an edge column into the graph as weight.
  • Pass --path-mode distance only if the weight column represents distance, cost, delay, or impedance.
  • Pass --path-mode strength only if the weight column is a normalized 0..1 relationship strength. Internally, the toolkit converts strength into a non-negative cost with -log(strength).
  • Do not call a route "fastest" unless an edge column truly encodes time or delay.

Outputs

Typical outputs:

  • graph_summary.json: node/edge counts, density, components, isolates.
  • node_features.csv: graph-derived node features.
  • nearest_targets.csv: nearest target node and shortest path per source.
  • shortest_paths.csv: source-target path table.
  • model_metrics.json: optional outcome-model metrics.
  • model_feature_importance.csv: optional RandomForest feature importances.

Minimum Example

python -m vahtian.epinet.toolkit \
  --nodes synthetic_nodes.csv \
  --edges synthetic_edges.csv \
  --outcome-column Outcome \
  --target-outcome 1