IOHinspector is a Python package designed for processing, analyzing, and visualizing benchmark data from iterative optimization heuristics (IOHs). Whether you're working with single-objective or multi-objective optimization problems, IOHinspector provides a collection of tools to gain insights into algorithm performance.
IOHinspector is a work-in-progress, and as such some features might be incomplete or have their call signatures changed or expanded when new updates release. As part of the IOHprofiler framework, our aim is to achieve feature parity with the IOHanalyzer web-version. Additionally, this package serves as the basis for large-scale data processing, which is currently unspported on the web version.
- Data Processing: Efficient import and process benchmark data. We currently only support the file structure from IOHexperimenter, but this will be expanded in future releases. By utlizing polars and the meta-data split, large sets of data can be handled efficiently.
- Analysis: Perform in-depth analyses of single- and multi-objective optimization results. For the multi-objective scenario, a variety of performance indicators are supported (hypervolume, igd+, R2, epsilon), each with the option to flexibly change reference points/set as required.
- Visualization: Create informative plots to better understand the optimization process. This included standard fixed-budget and fixed-target plots, EAF and ECDF visualization and more.
The minamal suported Python version is 3.10. Install IOHinspector via pip:
pip install iohinspectorThe basic usage of the framework is through the data manager object. A simple example is given below. This assumes that a folder created via the IOHexperimenter called test_data exists and contains profiling data. This can be generated via the file generate_test_data.py in the tests folder.
import os
from iohinspector import DataManager, plot_ecdf
# Creating a data manager
manager = DataManager()
data_folders = [os.path.join('test_data', x) for x in os.listdir('test_data')]
manager.add_folders(data_folders)
# Loading & selecting data
selection = manager.select(function_ids=[1], algorithms=['algorithm_A', 'algorithm_B'])
df = selection.load(monotonic=True, include_meta_data=True)
# Creating an ECDF plot
plot_ecdf(df)
To highlight the usage of IOHinspector, we have created two tutorials in the form of jupyter notebooks:
This project is licensed under a standard BSD-3 clause License. See the LICENSE file for details.
This work has been estabilished as a collaboration between:
- Diederick Vermetten
- Jeroen Rook
- Oliver L. Preuß
- Jacob de Nobel
- Carola Doerr
- Manuel López-Ibañez
- Heike Trautmann
- Thomas Bäck
Citation information coming soon!