Optimised Fitness Functions for runtime improvement , an application on MAGPIE

Introduction

We are integrating 13 measures and the retry parameter into MAGPIE and performing experiments on 7 benchmarks. We also performed consistency and correlation to rutnime experiments for the 13 measures + linus time.

Magpie is a tool for automated software improvement. It implements MAGPIE, using the genetic improvement methodology to traverse the search space of different software variants to find improved software.

Magpie provides support for improvement of both functional (automated bug fixing) and non-functional (e.g., execution time) properties of software.
Two types of language-agnostic source code representations are supported: line-by-line, and XML trees. For the latter we recommend the srcML tool with out-of-the-box support for C/C++/C# and Java.
Finally, Magpie also enables parameter tuning and algorithm configuration, both independently and concurrently of the source code search process.

Requirements

• Unix (Linux/macOS/etc; untested on Windows)
• Python 3.8+

Try it now!

git clone https://github.com/bloa/magpie.git
cd magpie
python3 magpie local_search --scenario examples/triangle-c/_magpie/scenario_slow.txt

Documentation

Everything you need to know about Magpie and the new Optimised Fitness Function and retry parameter.

Results

• Results for local search approach for 14 measure, retry combos on 11 scenarios
• Results for genetic programming approach for 14 measure, 5 retry combos on 11 scenarios

Optimised Fitness Functions

• Consistency - corellation experiments

• Guide to running a new benchmark

• Guide to create a scenario file

• Analysis Methodology

• Understanding the Results

• Test on Irace Tutorials

• Quick start (start here!)

• Search for improved variants

How-to guides

• Write a custom fitness function

Explanations

• Project structure

Reference guides

• Entry points
• Fitness functions
• Scenario files
• Algorithm configuration model

For ParamILS experiments

• Go to this folder

For LLM based experiments

• Go to this folder