Your first solve

Documentation index · CLI reference · Python API

Build the CLI, solve a small instance, and choose the result your application needs. Run every command below from the repository root. The example inputs are included in the checkout.

Build the CLI

Install Rust/Cargo, Git, and a C++14 compiler, then clone with submodules:

git clone --recurse-submodules https://github.com/lseman/baldes.git
cd baldes
cargo build --release --features clp
./target/release/baldes --help

The clp feature supplies the routing master and CLI. A default cargo build --release builds the core library. For optional backends and Make configuration, see Build and installation.

Choose a solve mode

Start with a small instance or customer subset. A time limit can stop a run before convergence; it does not promise an optimal solution.

Find a feasible solution

Run Hybrid Genetic Search on the bundled 32-node CVRP instance:

./target/release/baldes hgs instances/cvrplib/A-n32-k5.vrp --time 10

Use this mode when you need routes quickly. HGS searches for a feasible integer routing plan; its objective does not prove a lower bound or optimality.

Solve a root relaxation

Keep ten customers from a Solomon instance with time windows:

./target/release/baldes solve instances/Solomon/R101.txt \
  --customers 10 --time 60 --verbose --json root-result.json

The progress table goes to stderr; root-result.json stores the result for inspection. Look at the solve status, LP objective, selected bound, and route values. Root route values may be fractional, so these columns are not necessarily a dispatchable routing plan.

A root status of Optimal means the LP relaxation converged. It does not establish an optimal integer solution. The default bound mode is numerical; independent bound validation is available with --bound-mode certified. An unknown selected bound is null in JSON. See the CLI guide for the bound fields and validation behavior.

Search for an integer solution

Add --tree to explore the branch-and-price tree:

./target/release/baldes solve instances/Solomon/R101.txt \
  --customers 10 --tree --time 60 --verbose

An incumbent is a feasible integer solution. The lower bound and incumbent define the remaining gap. Tree Optimal means that gap closed within the configured tolerances. If the run reaches a limit, inspect its status and gap before treating the incumbent as optimal. See the tree guide for node limits, branching, and inherited state.

Keep distance conventions consistent

Routing commands default to --rounding tenth, truncating distances to tenths. Use --rounding nearest for TSPLIB nearest-integer distances, or --rounding none for unrounded Euclidean distances. Explicit distance matrices retain their supplied weights.

When comparing an objective with a published result, match the distance convention and customer count first. BALDES root benchmarks use --rounding none; see the benchmark setup.

Common setup fixes

Continue with your model

• Python API: construct customers in memory and inspect typed root results.
• CLI reference: cuts, logging, distance modes, and backend selection.
• Multi-pricing: coordinate multiple route families in one master.
• Options reference: supported pricing controls and their scope.