RouteOptimizer

RouteOptimizer CLI demo: scenario to dispatch and shift-plan JSON

CLI in and dispatch.json out

This repository has two parts:

Dispatch policy — a small neural network in dispatch_policy/ (13 → 16 → 8 → 1). It was trained to imitate the A* solver’s choice among legal dispatch actions on synthetic mines. On a held-out set it picks the solver’s action about 91% of the time (a demand-greedy baseline is about 56%). The checkpoint does not prove a plan is optimal.
Operations-research solvers — A* dispatch and a PuLP/CBC shift plan, which still produce the proven plans. Sample scenarios and the file format are included so you can run those solvers offline.

There is no haul-history model and no claim that the network replaces the solver.

Rights: Alfonso Bonillas / alioit.com. License: MIT (LICENSE).

Trained weights

File	Role
`dispatch_policy/weights.npz`	NumPy checkpoint (layer matrices `W0`…`W2`, biases, feature mean/std)
`dispatch_policy/config.json`	Architecture, feature names, holdout accuracy

from route_optimizer.policy import load_policy

policy = load_policy()  # packaged checkpoint, or pass a path to weights.npz
action = policy.choose(state)  # highest-scoring legal action

Teacher: A* on synthetic format-v1 scenarios. License: MIT.

What it does

Dispatch policy — ranks the legal moves at a fleet state.
Dispatch (A*) — builds a truck movement plan over a shift horizon from a scenario file.
Shift plan (MILP) — solves a mine-wide truck-flow plan with PuLP + CBC (open-source; no commercial solver license).

Both tools read local JSON scenarios and write JSON results. They do not call SQL Server or the network at run time.

What it does not do

It does not train or serve ML models.
It does not include production haul databases or live telematics.
It does not require Gurobi, CPLEX, or other commercial solvers.
A second anonymized haul-history sample is not published until the data owner approves release.

Install

pip package

pip install route-optimizer

Interactive demo: Hugging Face Space · Colab notebook.

Console scripts: route-optimizer, route-optimizer-dispatch, route-optimizer-shift-plan.

Docker image

Build from the same repository (image tag matches the package version in pyproject.toml, currently 0.1.0):

git clone https://github.com/AlfonsoALIO/RouteOptimizer.git
cd RouteOptimizer
./scripts/build_docker.sh
# → route-optimizer:0.1.0

Run with a mounted folder and networking disabled:

docker run --rm --network=none \
  -v "$PWD/samples:/data/samples:ro" \
  -v "$PWD/out:/data/out" \
  route-optimizer:0.1.0 \
  dispatch /data/samples/tiny_feasible.json -o /data/out/dispatch.json --segments 8 --trucks 2

Scenario file format

See docs/scenario-format.md (format version "1"). Units: tons, hours, minutes.

Sample scenarios

File	Role
`samples/tiny_feasible.json`	Small mine; dispatch and shift-plan both succeed quickly
`samples/toy_mine.json`	Larger synthetic toy mine used by the in-repo fixture (dispatch)

# pip — end-to-end on the small sample
route-optimizer dispatch samples/tiny_feasible.json -o out/dispatch.json --segments 8 --trucks 2
route-optimizer shift-plan samples/tiny_feasible.json -o out/shift_plan.json --segments 4

# optional wall-clock cap
route-optimizer dispatch samples/toy_mine.json -o out/toy_dispatch.json --segments 58 --trucks 21 --time-limit 60

Links

Resource	URL
Source / pip install	https://github.com/AlfonsoALIO/RouteOptimizer
Docker build script	https://github.com/AlfonsoALIO/RouteOptimizer/blob/master/scripts/build_docker.sh
Scenario format	`docs/scenario-format.md`
Synthetic scenarios (dataset)	https://huggingface.co/datasets/albones80/route-optimizer-scenarios
License	`LICENSE` (MIT)
alioit.com	https://www.alioit.com/route-optimizer

Citation / credit

RouteOptimizer — Alfonso Bonillas / alioit.com — MIT License.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

albones80
/

route-optimizer