# B5_MAX3SAT

**Repository Path**: beleton/b5_max3sat

## Basic Information

- **Project Name**: B5_MAX3SAT
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-02-02
- **Last Updated**: 2026-02-02

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# MAX-3-SAT (B5) Comparative Analysis Project

This repo contains code to generate MAX-3-SAT instances (DIMACS CNF), run 3 classes of algorithms, and collect comparable experiment results.

Environment (per course spec):

```bash
conda activate math
```

## Quick start (sanity run)

Generate a tiny dataset and run a benchmark:

```bash
python -m src.experiments.generate_instances \
  --out-dir data/instances \
  --meta-dir data/meta \
  --categories random_uniform planted biased_polarity community near_threshold \
  --scales small \
  --per-class 5 \
  --seed 0

python -m src.experiments.run_benchmark \
  --instances-root data/instances \
  --results-dir data/results \
  --algorithms exact_bnb random walksat \
  --repeats 5 \
  --seed 0

python -m src.experiments.stats \
  --results-csv data/results/latest/results.csv \
  --out-dir data/results/latest
```

## Structure

- `src/`: implementation
  - `src/cnf.py`: DIMACS CNF I/O and CNF data model
  - `src/evaluate.py`: evaluation + incremental flip bookkeeping
  - `src/algorithms/`: three algorithm families
  - `src/experiments/`: instance generation, benchmarking, stats
- `data/instances/`: generated `.cnf` instances
- `data/meta/`: instance metadata JSON
- `data/results/`: raw results CSV/JSONL and summaries

## Notes

- Plotting is optional; if you want plots, install `matplotlib`.
- The exact solver is intended for small instances only.