v1.1.0

In this release, we added support for regression tasks and MoleculeNet datasets used in Tumescheit et al.: Ontology pre-training improves machine learning-based predictions for metabolites (bioRXiv, 2025) (#130). In the context of our new ensemble (Flügel et al.: Chebifier 2: An ensemble for chemistry (SymGenAI4Sci Workshop, 2025)), we now have SMILES canonicalisation by default (#118) and support Logistic Regression and LSTMs (#127). See below for a full list of changes.

What's Changed

• Classwise F1-scores to generate class props script by @aditya0by0 in #112
• add smiles canonicalisation, update tokens.txt by @sfluegel05 in #118
• allow negative samples (and regulate their amount for partial data) by @sfluegel05 in #116
• Fix data splits for pubchem by @sfluegel05 in #119
• Test for CLI + project dependencies changes by @aditya0by0 in #105
• split dependencies into inference (mandatory), training and linters by @aditya0by0 in #114
• Data Augmentation : SMILES by @aditya0by0 in #115
• Enable to set persistent_workers through CLI by @aditya0by0 in #126
• Avoid using iterrows, use vectorization wherever possible by @aditya0by0 in #120
• Fix issue in mlp architecture by @aditya0by0 in #128
• Fix: evaluation skips last batch by @jcapp4 in #129
• Feature/new ensemble models by @sfluegel05 in #127
• Fix for weighted BCE loss by @aditya0by0 in #132
• Enable to load pretrained weights for MLP model by @aditya0by0 in #133
• Avoid generation of original SMILES in augmentation by @aditya0by0 in #136
• New regression and classification datasets for ontology pre-training by @schnamo in #130

New Contributors

• @jcapp4 made their first contribution in #129

Full Changelog: v1.0.3...v1.1.0

v1.0.3

This release includes some minor fixes:

What's Changed

• fix import for XYBaseDataModule by @sfluegel05 in #107
• Upgrade pyproject backend by @aditya0by0 in #109
• fix handling of 0-label predictions by @sfluegel05 in #110

Full Changelog: v1.0.2...v1.0.3

v1.0.2

This release includes:

• Change from setup.py to pyproject.toml - we also took the opportunity to remove some imports that are no longer needed and move some rarely used imports to extras. This should speed up installation #99
• Separate test and validation split ratios (with updated default values: test set is now 10%, validation set 5% (compared to ~13% and ~2%)) #103
• Fixes: hyperparameters get saved corrects #97, trainer is now determinstic #101, minor fixes for ensemble #104

v1.0.1

This release includes:

• Support for using chebai-trained models in chebifier ensembles (e.g., via automatically calculating metrics on the validation set) #102
• Fixes and improvements related to the chebai-extensions in chebai-graph and chebai-proteins #96 #93 #91 #88
• Streamlining the load_processed_data function #92 #90
• Some minor fixes and improvements #84 #85

v1.0.0

Changes in the last months include:

• #80 Protein datasets have been moved to python-chebai-proteins
• #79 n_token_limit allows pruning dataset to predetermined length
• #74 model.out_dim does not have to be given as a command line parameter but is inferred from the dataset
• #69 / #71 The fuzzy loss from Flügel et al.: A fuzzy loss for ontology classification, NeSy 2024 has been added
• improved documentation, unittests and workflows

v0.0.0

Merge remote-tracking branch 'origin/master'