Changelog¶

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

[Unreleased]¶

runelog serve command to deploy modfels as a local API
Lightweight feature store implementation
More visualizations options, i.e. ROC curves, feature importances, confusion matrices
Extensible plugin architecture for custom trackers or visualizations

Git Integration: Automatically logs the Git commit hash, branch name, and repository "dirty" status for every run to a source_control.json file.
Environment Snapshot: Automatically logs the Python version, platform information, and a full list of installed packages to an environment.json file. It also creates a pip-installable requirements.txt as a run artifact.
Automatic Code Logging: Now includes an opt-in feature in start_run to automatically save a copy of the executing script as an artifact.
Data Versioning:
- Data Hashing: Added tracker.log_dataset(), a new method that calculates a SHA256 hash of a data file and saves it to a data_meta.json file for verifiable data lineage.
- DVC Integration: Added tracker.log_dvc_input(), a helper that reads the hash from a .dvc file to link runs with DVC-managed data.
Run Lineage: Added tracker.log_input_run(), which creates a lineage.json file to explicitly link a run to its upstream dependencies, such as a feature generation run.
Documentation: Added a new guide explaining the "Lightweight Feature Store" pattern and a guide for the new DVC integration.

The start_run method now accepts an experiment_name directly, simplifying the most common user workflow.
Added delete_run method to the core library for better cleanup and management.
Added runs delete command with interactive confirmation prompts for safety.
Improved empty state of the Streamlit UI for fresh runs, showing a brief quickstart when no experiments exist yet.

Corrected the development installation instructions in the README.md and contribution guides.

Refactored pyproject.toml to use a pure hatchling build backend and a new Hatch environments for docs.
Changed usage examples in examples/ to reflect API changes.

Experiment Tracking: RuneLog class for managing experiments and runs. Supports logging parameters, metrics, artifacts, and models.
Model Registry: Full-featured model registry with versioning and tagging.
Sweep Runner: run_sweep function for automated experiments from a flexible YAML configuration file.
Custom Exceptions: A full suite of specific exceptions for robust error handling.

A full-featured CLI powered by Typer and rich.
runelog experiments: list, get, delete or export experiments to CSV.
runelog runs: list, get, compare runs side-by-side, and download-artifact.
runelog registry: list models, get-versions, register a model, and tag versions.
runelog sweep: Execute a sweep from a config file.
runelog ui: Launch the web UI.
runelog examples: Commands to run example scripts.

Experiment Explorer: View experiments and runs with a detailed drill-down view.
Visual Run Comparison: Select multiple runs to see an interactive bar chart comparing their performance.
Artifact Previewer: Render common artifact types like images and text files directly in the UI.
Model Registry Viewer: Browse registered models and their versions.
Register from UI: A button in the run detail view to register a model directly.

Professional Project Structure: Uses a src-layout managed by Hatch.
Testing: Comprehensive test suite using pytest, including unit and integration tests.
Docker Support: Dockerfile and docker-compose.yml to easily build and share the UI.
Documentation: A full documentation site built with mkdocs.
Community Files: LICENSE, CONTRIBUTING.md, and CODE_OF_CONDUCT.md.