Changelog

All notable changes to querygym will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

0.2.0 - 2026-04-12

Added

• ThinkQE reformulation method: multi-round query expansion via iterative corpus feedback
• Strips <think>...</think> reasoning traces for compatibility with DeepSeek-R1 style models
• Configurable rounds (num_interaction), expansion accumulation, and novelty-based passage filtering
• Prompt template thinkqe.v1 added to prompt bank
• Full documentation in methods reference and reformulation guide
• ThinkQE listed in README supported methods table

Fixed

• Removed misspelled reqeat_weight fallback key in ThinkQE._get_repeat_weight()
• Renamed test_method_thinkqe.py to test_methods_thinkqe.py for naming consistency

0.1.4 - 2025-11-21

Added

• Docker support with pre-built images on GitHub Container Registry (GHCR)
• GPU-enabled image (ghcr.io/ls3-lab/querygym:latest)
• CPU-only image (ghcr.io/ls3-lab/querygym:cpu)
• Multi-platform support (linux/amd64, linux/arm64 for CPU)
• Docker Compose configuration for easy setup
• Comprehensive Docker documentation and examples
• GitHub Actions workflow for automated Docker image building and publishing
• Example notebooks and scripts in examples/ directory
• Code snippets for quick reference
• Docker examples with Jupyter notebooks
• QueryGym + Pyserini integration examples
• DOCKER_SETUP.md quick reference guide
• Makefile for local Docker development

Changed

• License changed from MIT to Apache License 2.0
• Reorganized examples into three categories: snippets, docker, querygym_pyserini
• Updated to OpenJDK 21 (from 17) in Docker images
• Migrated to Docker Compose V2 syntax (docker compose instead of docker-compose)
• Improved Docker image tagging strategy (separate tags for GPU and CPU)
• Enhanced README with Docker installation options and citation information
• Updated all documentation to use modern Docker Compose commands

Fixed

• Docker build issues with OpenJDK availability in Debian Trixie
• Docker Compose tag conflicts between GPU and CPU images
• Inconsistent command examples across documentation

0.1.3 - 2025-11-19

Added

• Apache License 2.0
• Citation information with arXiv paper reference in README

Changed

• Package metadata updated with Apache 2.0 license

0.1.2 - 2025-11-18

Changed

• Updated author information with email addresses for Mert Incesu and Negar Arabzadeh
• Version bump for metadata updates

0.1.1 - 2025-11-18

Added

• Added tqdm as core dependency for progress bars
• Complete Read the Docs documentation setup
• MkDocs configuration with Material theme
• Comprehensive API reference documentation
• GitHub Actions workflow for automated PyPI publishing

Changed

• Updated package name from queryGym to querygym (lowercase)
• Updated author information in package metadata
• Improved dependency management

Fixed

• Fixed missing tqdm dependency in core package

0.1.0 - 2025-11-17

Added

• Initial release of querygym
• Core reformulation framework
• Eight query reformulation methods:
• GenQR (generic keyword expansion)
• GenQR Ensemble
• Query2Doc
• QA Expand
• MuGI
• LameR
• Query2E
• CSQE
• Prompt bank system with YAML configuration
• DataLoader for queries, qrels, and contexts
• BEIR and MS MARCO format loaders
• Pyserini and PyTerrier adapters
• CLI interface with querygym command
• OpenAI-compatible LLM client
• Comprehensive test suite
• Example scripts and documentation

Features

• Dependency-free core data loading
• Optional integrations (BEIR, Pyserini, HuggingFace)
• Flexible LLM client supporting any OpenAI-compatible API
• Batch processing with progress bars
• Context-based reformulation support
• Export to TSV/JSONL formats
• Script generation for Pyserini + trec_eval