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.
- Revert to linear calibration after transfer learning only
- Use scikit-learn instead of pygam for calibration
- Fix custom activation by using default implementation
- Loosen requirements
- Remove batch number parameter from GUI
- Debugging release issues
- Debugging release issues
- Debugging release issues
- Remove old python build
- Debugging release issues
- Windows release fix
- New TensorFlow versions that break support with earlier models
- Relax tensorflow version for mac
- Pin scipy version
- Set max threads
- Remove limit threads feature calc
- Fix issue transfer learning single model mode
- Make single model mode available and the default
- Bump version
- Bump version
- Bump version with new workflows
- Bump version with .toml
- Removed dependencies (most importantly sklearn; which is optional now)
- Support for custom labels in plot
- Fix memory usage, limit threads
- Fixed multiprocessing
- Pass flag CCS feature extract
- Reintroduce ability to predict CCS
- Fix pypi token GA
- Fix non-initialized modifications
- Allow for setting number of epochs
- Fix selenium atom logging error
- Fix selenium atom on AA
- Fix wrong error except
- Fix peptides that are too long and their modifcations
- bump version
- reduce logging (change to debug)
- Fix missing debug time in feature extractor
- Fix potential issues in feat extractor (uncommon AA)
- Set plotly param to false
- Remove assumption psm_utils_obj
- Add plotly diagnostic plots
- Fix logger issue
- Activate garbage collection to clear GPU memory
- Reintroduce support for batched calculation
- Introduce batch_num_tf parameter for DeepLC that determines TF batch size
- Allow for linear piecewise calibration again
- More strict version of TF, as 2.13.0 crashes for transfer learning
- Fixed a bug where a reinit led to issues with setting parallelism
- Fixed a bug where it checked the calibration file while this was not initialized
- Fixed a bug where missing atoms in encoding cause a crash
- Fixed a bug where isotopes are incorrectly parsed
- Fixed a bug where numpy no longer accepts dict_values, explicit list conversion
- Fixed a bug where atom counts were wrong, fixed by retraining models with new features
- Fixed a bug where detection of legacy csv is wrong
- remove pygam from GUI, set to true as default
- fix setting cmd line calibration default
- fix setting GUI calibration default
- fix library feature
- slight refractoring
- slight refractoring
- add Arthur Declercq as contributor
- dependency fix
- Calibration bug fix
- Support for GUI and PSM utils
- Support for GUI and transfer learning
- Log peptide sequences that are too long
- bump version
- second try
- bump version
- bug fix for sequences > 60 AA
- bug fix non-resolved modifications
- psm_utils integration
- pygam default calibrator
- additional support transfer learning
- code refractoring of feature extraction
- prediction library support
- piecewise linear calibration support
- Bump version
- Support for DeepLCRetrainer
- Optimized library usage (~85% reduce in DeepLC prediction time) @markmipt
- Made Gooey an optional dependency (facilitates install on Linux). Install the optional
dependencies for the graphical user interface with
pip install deeplc[gui]
- New native Python GUI, based on the Gooey package
- New standalone installer for Windows using PyInstaller and Inno Setup
- Added
deeplc-gui
entrypoint to start GUI from the command line
- CLI: Restructured help message
- Made DeepLC class API docstring consistent with CLI help message
- Docs: Moved
dict_divider
andsplit_cal
explanation to README Q&A section. - CI: Only run tests on commits to
master
or from a PR - Refactoring: Cleaned up
__main__.py
- Logging: Changed some loggings from DEBUG to INFO level, some from WARNING to INFO or DEBUG level
- Removed Java-based GUI in favor of new Python-based GUI
- If run through CLI/GUI, all Tensorflow warnings are now fully suppressed
- Added
--legacy_calibration
CLI option to allow for old piecewise linear calibration while new pyGAM calibration method is default.--legacy_calibration
is mutually exclusive with--pygam_calibration
.
- Make version compatible with pip release
- Make version compatible with pip release
- Make pygam the default calibration method
- Bug fix where split_cal was not correctly passed
- CMD support for pygam calibration
- Scikit-learn and pygam dependency in setup
- Bug fix duplicate peptide+mod for DeepCALLC
- New feature DeepCALLC
- Version bump
- Deep(CAL)LC functionality
- Pygam as calibration function
- Update to Streamlit webserver: Use
st.form
and new official download button
- More elegant solution for library call as global
- Fix var call dict object
- Fix library delete call
- Temporary fix suggested by markmipt for the library
- Change logging to specified logger object instead of standard logger
- GUI: Fix small font through starting jar with cmd to increase font size
- Added testing for Python 3.9
- Relax h5py requirement to allow v3
- Fixed GitHub Action workflow for Streamlit docker image build
- Bug in writing library where a list was assumed so library only partially filled
- Make it optional to reload library
- Ignore library messages
- Force older library h5py for compatability
- Library gets appended for non-calibration peptides too
- Change the default windows install to pip instead of conda
- Change reporting of identifiers used from library
- Log the amount of identifiers in library used
- Publish PyPI and GitHub release
- Add library functionality that allows for storing and retrieving predictions (without running the model)
- Describe hyperparameters and limit CPU threads
- Additional modfications, including those exclusive to pFind
- Change calibration error to warning (since it is a warning if it is out of range...)
- Limit CPU usage by tensorflow by connecting to n_jobs
- Support for Python 3.8
- Bug fix in the calibration function
- Had to order the predicted instead of observed retention times of the calibration analytes
- Thanks to @courcelm for both finding and fixing the issue
- Different calibration function, should not contain gaps anymore
- Changed to more accurate rounding
- Changed to splitting in groups of retention time instead of groups of peptides
- Changed default model to a different data set
- Duplicate peptides charge fix
- Support for charges and spaces in peprec
- Fixes in GUI
- Include less models in package to meet PyPI 60MB size limitation
- Bugfix: Pass custom activation function
- Fixed support for averaging predictions of groups of models (ensemble) when no models were passed
- New models for ensemble
- Support for averaging predictions of groups of models (ensemble)
- Fix the latest release
- Spaces in paths to files and installation allowed
- References to other CompOmics tools removed in GUI
- Fixes in GUI
- Fix the latest release
- Fixed the .bat installer (now uses bioconda)
- Example files in GUI folder
- Unnecesary bat and sh for running GUI removed
- Switch to setuptools
- Reorder publish workflow; build wheels
- Remove xgboost dependancy
- Use dot instead of dash in versioning for bioconda
- Fix publish action branch specification (2)
- Fix publish action branch specification
- Test other trigger for publish action
- Update documentation, specify branch in publish action
- Add long description to setup.py
- Initial pre-release