Skip to main content
Open llms.txtCopy tools

Lane-Level Acceptance

Lane-level checks exercise combined dense, temporal, spatial, and sparse route-cell behavior on deterministic synthetic data.

Command​

Install from PyPI for normal use:

uv add cartoboost

From a source checkout, rebuild the local native extension before regenerating the maintained acceptance artifacts:

uv run --group dev maturin develop
uv run --group dev python scripts/run_lane_level_acceptance_metrics.py

Dataset Shape​

  • 4 origin regions x 4 destination regions = 16 lanes.
  • 24 hourly observations per lane.
  • Observable columns: origin x/y, destination x/y, lane ID, hour, midpoint x/y, and distance.
  • No hidden simulator metadata is passed into training.

Outputs​

Generated outputs live under docs/assets/lane_level_tests/:

  • acceptance_metrics.json
  • acceptance_metrics.md
  • route_midpoint_cartometry.png
  • hour_profile.png
  • lane_heatmap.png

Use these files to inspect route cartometry, hour effects, lane-level residuals, and combined split behavior.

What The Check Proves​

The lane-level dataset is intended to show whether CartoBoost captures:

  • Route-cell sparse-set encoding.
  • Temporal profile behavior.
  • Spatial route cartometry behavior.
  • Combined split behavior when several feature families are present.

It is not a production quality benchmark. Any broader benchmark claim needs a documented dataset, feature handling, baseline configuration, and repeated-run summary.