optimaze 0.3.0

AI-powered optimization model tuning

Gurobi Optimizer Agent

Automatically speed up any Gurobi optimization model.

An AI agent that analyzes your model, identifies bottlenecks, and finds the best solver settings and formulation improvements.

What It Does

Input: Your Gurobi model (.mps, .lp, or gurobipy)
Output: Faster solve time + recommendations

┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│  Your Model │ →  │   Agent     │ →  │  30% faster │
│             │    │  analyzes   │    │  + report   │
└─────────────┘    └─────────────┘    └─────────────┘

Quick Start

# Install
pip install gurobipy numpy

# Clone
git clone https://github.com/danielpuri1901/tax-optimization-agent.git
cd tax-optimization-agent

# Run on your model
python demo.py

Usage

from src.agent.general import GurobiAgent

# Load your model
import gurobipy as gp
model = gp.read("your_model.mps")  # or build with gurobipy

# Run the agent
agent = GurobiAgent()
result = agent.improve(model)

# See results
print(f"Best: {result.best_improvement} ({result.best_speedup:+.1f}%)")
print(f"Params: {result.best_params}")

How It Works

1. Profile - Analyzes model structure (variables, constraints, sparsity, symmetry)
2. Match - Selects improvements likely to help based on detected patterns
3. Smoke Test - Quick tests to filter out bad candidates
4. Benchmark - Full tests on promising improvements
5. Report - Returns best settings and recommendations

Example Results

Model	Baseline	Best Improvement	Speedup
Supply Chain (150K vars)	0.194s	disaggregate_constraints	+23.7%
Tax Optimization (10K constrs)	4.2s	aggregation	+99.8%
Facility Location (2K binary)	0.084s	Already optimal	-

What It Tries

Solver Parameters:

• MIPFocus (feasibility, optimality, bound)
• Cuts (off, moderate, aggressive)
• Presolve (off, conservative, aggressive)
• Method (primal, dual, barrier)
• Heuristics, Threads, Symmetry

Formulation Transforms:

• Variable bound tightening
• Symmetry breaking constraints
• Constraint disaggregation

Requirements

• Python 3.11+
• Gurobi with valid license (free academic license)
• numpy

Project Structure

├── src/agent/general/     # Core agent
│   ├── agent.py           # Main GurobiAgent class
│   ├── profiler.py        # Model analysis
│   ├── catalog.py         # Improvement library
│   └── matcher.py         # Rule-based selection
├── demo.py                # Simple demo
├── demo_supply_chain.py   # Supply chain example
└── run_general_agent.py   # Full demo with multiple models

License

MIT - see LICENSE

Contributing

Issues and PRs welcome. If you have a model where the agent helped (or didn't), we'd love to hear about it.