Reference

Data & Inputs

Check that GET can represent your network, then convert external identifiers into the integer node indices the engine uses.

Is GET Suitable for This Network?

GET currently represents	Implication
A fixed number of nodes	Evolution changes edges, not the node set.
Undirected edges	A regulatory edge `A → B` cannot retain its direction.
Integer edge multiplicities	Continuous weights, signs, and confidence scores need an explicit conversion.
No node or edge attributes	Gene type, pathway, condition, and other metadata must live in your objective or a side table.
Dense graph storage	Memory grows quadratically with node count; benchmark a representative small run before committing to a large network.

Do not silently discard domain meaning to fit the model. If direction, signed edges, attributes, or a changing node set are central to the research question, GET's current graph type is not a faithful representation without a reviewed transformation.

Edge-file Format

Each file is comma-separated, contains one integer start,end,weight edge per line, and includes exactly one node-count header. Lines beginning with # are comments.

# nodes = 5
# source IDs are recorded in node_ids.csv
1,2,1
2,4,1
4,5,1

Pass min_node_index=1 for this 1-based example; use 0 for 0-based data.
Self-loops, non-integers, out-of-range nodes, negative weights, and weights above the configured cap are errors.
Undirected duplicates such as 2,4 and 4,2 collapse; the last weight wins with a warning.
A zero-weight row adds no edge and produces a warning.
The header is required because isolated nodes cannot be inferred from edge rows.

evolver.set_base_graph_from_file("base_graph.csv", min_node_index=1)

Choose How the Base Graph Is Loaded

For the usual TOML route, put the path directly under an edge-edit genome. Relative paths resolve from the TOML file's directory, not from the shell's working directory:

[genome]
type = "edge_edit"
gene_length = 256
base_graph = "base_graph.csv"

This config key reads 0-indexed nodes and requires the file's # nodes = N value to equal network_size. Use set_base_graph_from_file instead for a 1-indexed file, a dynamically chosen file, or a configuration assembled as Python objects. Use one loader per evolver; do not apply both to the same run.

Preserve Biological Identifiers

GET returns the numbering supplied to the first graph loader, but it does not store gene or protein labels. Create and archive a one-to-one mapping before the run:

node_id,biological_id
1,TP53
2,MDM2
3,CDKN1A
4,ATM
5,BAX

Normalize identifiers to one namespace and record the source/database version.
Assign each retained identifier one integer; keep isolated nodes in the mapping and node count.
Build the edge file using those integers and archive both files together.
Map result.best_edges back through the same table; never infer labels from row order later.

Reference Graph Folders

The struct_match objective reads every usable reference graph from reference_folder. Each graph uses the same edge-file format and its own # nodes = N header. The folder must exist and contain at least one usable graph. Archive the exact folder contents and identifier conventions; the TOML path alone does not preserve them.

Provenance Checklist

GET tag and resolved commit, plus dependency/environment details
Run code and complete configuration
Master seed, replicate index, and any custom-objective seed
Base graph, reference folder, identifier map, source/version, and licenses
Custom objective code and scientific reporting/validation method

Continue to Configuration Reference for parameter details, Python with TOML for the normal config-file route, or Python config objects for explicit loaders.