Q&A
1 Solver Loading & Folder Name
The scoring system uses the same loading mechanism as the published 2_run.py:
Copy my_solver/ and rename it to your team name, then run with --solver <team_name>. Other folders in the repository (e.g. my_solver) will not affect scoring.
The Guide page has also been updated to reflect this workflow.
2 Python / Package Versions
The scoring environment runs Python 3.12. Packages like numpy and pandas are installed via pip install -r requirements.txt with no specific version constraints.
3 Runtime Environment
The scoring server is CPU-only; GPU is not available.
In real AOI equipment, GPU and other resources are dedicated to Inspection and Reconstruction tasks. Planning runs on CPU. The contest environment reflects this real-world setup.
All teams run on the same server, ensuring no hardware-based advantage. Datasets are executed sequentially in a single process. Teams are free to use multiprocessing within their solver.
4 Computation Time Measurement
Only the solve() call is timed (package import time is excluded):
5 Time Limit
The time limit is a safety measure to prevent excessive iterative computation and random search from overloading the scoring system. The exact limit is determined by the scoring environment.
Additionally, since Computation Time is directly included in the score, longer computation naturally results in a worse score.
6 checker.py / 2_run.py Consistency
The utils/checker.py and 2_run.py used for scoring are identical to the versions published in the repository. Any changes will be announced in advance.
7 Hidden Dataset (Contest Test Set)
The hidden test set is generated from 1_generate_data.py within the parameter ranges described in the Data Description. No inputs outside those ranges (e.g., extreme component counts, side values other than 0/1) will be included.
8 Fiducial FOV Handling Updated
checker.py automatically handles Fiducial FOVs:
- FOV center is snapped to the fiducial component's center coordinates
- Non-fiducial components in the same FOV are separated; only the fiducial remains
Separated components are treated as uncovered, and penalty_solver is automatically applied.
(Updated 2026-09-19, contest repository & team private repositories)
9 Indirect Use of ky_solver
Benchmark comparison purposes (score comparison, output statistics analysis) are within the allowed scope.
The Guide states: "Use for purposes other than directly generating solutions is permitted." As long as ky_solver output is not used as input or solution generation for your own solver, it is fine.
10 Submission Method
The latest commit on the main branch at the deadline is the scoring target. No separate tags or releases are required.
Modifications after the deadline will not be reflected in scoring.