ODIN - Spectrum-based Fault Localization Tool

Artifact Information

This repository is the official artifact and source code of the tool ODIN, accepted at the SAC 2026 Conference.

Paper Title: An Empirical Study on Spectrum Based Fault Localization Heuristics

1. Enviroment Setup :

• Python version 3.11 is required
• Dependencies are included in requirements.txt file
• Install using: ./install_dependecies.sh
• For other Python versions, manually install:
  • matplotlib
  • numpy
  • pandas
  • scikit-learn
  • scipy
  • seaborn

2. How to Run The Experiments:

We have provided shell scripts to run all the experiments and generate the experimental results provided in the paper.

2a. Using Automation Script

python3 main.py --paper

This will utilize the spectrums that we sampled for our paper and perform the experiments. All experimental results will be available in ./paper/. The directory './paper/' contains all figures and data in their respective RQ folder. for example, if you want to see the results of RQ1 from the paper, please navigate to 'paper/experiment_type/RQ1', it will contain the images related to RQ1.

2b. Re-running Full Experiment from scratch

python3 main.py --resimulate --outputpath ./resimulated_results/

If you are interested in running the ODIN sampler and sampling fresh spectrums using all the configurations specified in Section 5 of the paper, please invoke the above given command. Note that, resampling the spectrums and generating the experimental results using this script takes around 10 hours in our system which has AMD Ryzen 9 9950X 16-Core Processor, with 16 cores and 32 threads @5.7GHz and 32 GB of RAM. All experimental results will be available in the specified output folder given in the command line argument.

3. Using ODIN Sampler

Use sampler.py with the following parameters:

Parameter	Description	Default Value
-n	Number of spectrums	1
-c	Components per spectrum	100
-b	Buggy components	1
-t	Test cases	10
-g	Flakiness (false-fail probability)	0.0
-d	Bug-triggering probabilities	[1.0]
-o	Output path	"./generated_spectrums/"

Example Usage

python3 sampler.py -n 10 -c 100 -b 2 -t 20 -g 0.0 -d [1.0,1.0] -o "TEST_SPECTRUMS/"

This generates 10 spectrums with:

• 100 components each
• 20 test cases
• 2 buggy components
• No false fails
• Output saved to "TEST_SPECTRUMS/"

4. Reference System Specifications

• Processor: AMD Ryzen 9 9950X 16-Core Processor
• RAM: 32GB

5. Minimum System Requirements

• RAM: 8GB
• Processor: Multi-core recommended
• Storage: 12GB minimum

6. Experimental Settings

• Component sizes: 1000, 5000, and 10000
• Ranks of buggy components are saved (~4GB) instead of full spectrums (>300GB)
• Generates synth_clusters.json in the provided output directory/folder for plot generation (RQs)

7. Project Structure

Scripts

• install_dependecies.sh: Installs Python dependencies
• RQ1.py, RQ2.py, RQ3.py: Research question analysis scripts
• SpectrumGenerator.py: ODIN sampler implementation
• metrics.py: Contains all 37 metrics used in the paper
• sampler.py: Spectrum sampling utility
• run_main_single.py: The script for sampling spectrums with 100% coverage as per prior single bug theoretical analysis
• run_main_multi.py: The script for sampling spectrums with 100% coverage as per prior multi bug theoretical analysis under Sx and Sz constraint
• run_main_multi_relaxed.py: This script relaxes only Sx and Sz constraint of run_main_multi.py and samples spectrums.
• frameworkXieEtAl.py: contains utlity functions which contains the preassumptions of prior theoretical works for generating and validating new spectrums.

Configuration Files

• config1.py: Main configuration
• config_theoritical.py: Theoretical sampling parameters
• config_single.py: Single bug analysis parameters under prefect detection scenario
• config_multi.py: Multi-bug analysis parameters under contrained Sx and Sz
• config_multi_relaxed.py: Multi-bug analysis parameters
• config_user.py: Custom experiment configuration

Data Directory Structure

data_experiment/
├── output_multi/
│   └── synthetic_clusters.json
├── output_multi_relaxed/
│   └── synthetic_clusters.json
├── output_single/
│   └── synthetic_clusters.json
└── outputs_theoritical/
    └── synthetic_clusters.json

Each synthetic_clusters.json contains pre-sampled ranks of buggy components for different control parameters.