Changelog

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

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

Updated code examples in docstrings and documentation to reflect current API
Optimized PSM list filtering to use shallow copy instead of full deepcopy for better memory efficiency on large datasets

CLI argument passing to predict function when calibration data is not provided

New core module with predict and predict_and_calibrate as main public API
New calibration module replacing calibrate, with Calibration and LinearCCSCalibration classes
New constants module for default model paths and configuration
New exceptions module (public, replacing private _exceptions)
New _model_ops module for PyTorch model loading and inference
New _io_helpers module (split from utils) for file parsing and I/O
Architecture subpackage (_architectures) with modular components: activations, blocks, callbacks, losses, helpers
Multi-conformer CCS prediction support (IM2DeepMulti)
PyTorch Lightning-based model architectures replacing Keras models
num_threads argument to control Torch CPU parallelization
Profiling support in CLI (--profile flag)
Test suite with tests for calibration, CLI, constants, core, exceptions, integration, losses, model ops, and utils
CI workflow for tests and type checking
This CHANGELOG file

Breaking: Public API changed from predict_ccs/linear_calibration to predict/predict_and_calibrate
Breaking: Minimum Python version raised from 3.10 to 3.11
Breaking: Switched from Keras/TensorFlow models to PyTorch Lightning models (.ckpt)
Switched build system from setuptools to uv
CLI restructured as click command group with predict as default command
Renamed reference data files for clarity
Linting moved from black/isort to ruff
Upgrade deeplc dependency to deeplc>=4.0.0b1,<5
Added explicit lower/upper version bounds to all core dependencies.