Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

38 Commits
 
 
 
 
 
 

Repository files navigation

Understanding APR Agents

Supplementary materials for analyzing Automated Program Repair (APR) agents on SWE-bench.

Environment Setup

# Install Python 3.9.0 (using pyenv)
pyenv install 3.9.0
pyenv local 3.9.0

# Create and activate virtual environment
python -m venv venv
source venv/bin/activate

# Install dependencies
pip install --upgrade pip
pip install -r requirements.txt

# Set up API keys
export ANTHROPIC_API_KEY="your-api-key-here"
export OPENAI_API_KEY="your-api-key-here"

Trajectory Data

Agent trajectories are available on HuggingFace:

Using Git LFS

# Make sure git-lfs is installed
git lfs install

# Clone the repository
git clone https://huggingface.co/datasets/zt6c3mxv8q/logs_and_trajs

Results

Tables (Markdown)

File Description
T1_solve_rates.md Solve rates by agent, difficulty, and backbone
T2_unique_counts.md Unique solution counts per agent (Venn analysis)
T3_agent_strengths.md Agent unique strengths by bug type (LLM-derived)
T4_clone_detection_similarity.md Clone detection and patch similarity by agent/difficulty
T5_gpt4o_hunk_stats.md Hunk-level patch statistics (GPT-4o backbone)
T6_overfit_rates.md Overfitting rates by agent and difficulty
T7_pass_rate_by_tests.md Pass rates stratified by number of tests
T8_cohens_k.md Cohen's kappa validation for clone classification
T9_error_diversity.md Error diversity across temperatures

Figures (PDF)

File Description
RQ1_motivating_ex.pdf RQ1 motivating example
rq1_patch_analysis_full_table.pdf Full patch analysis table
nebius-50-annotations.pdf Nebius-50 manual annotation results
localization_full.pdf Full localization analysis
samp_vs_temp.pdf Sampling vs temperature analysis

Detailed Documentation

Analysis README
Full pipeline instructions submission/README.md
Reproducer analysis (overfitting, issue reproduction) submission/code/reproducer_analysis/README.md
Reproduction test localization (inference scaling) submission/code/reproduction_test_localization/README.md
Error diversity analysis (temperature effects) submission/code/reproduction_test_localization/reproduction_test_error_diversity/README.md
Patch analysis (hunk stats, edit patterns) submission/code/patch_analysis_edit_hunks/README.md

Raw Data

About

[ISSTA '26] Understanding APR Agents Through the Lens of Traceability

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages