Changelog

Version 2.0.0.dev2 - 2025-04-06

This section is going to be updated and will become v2.0.0 eventually.

This is the first release with breaking changes, coming sooner than we would have liked.

The need for this came from relying on some legacy interfaces in torchtext for too long, which now have been dropped. The result was that with never PyTorch Versions, which also requires a matching torchtext version, which in turn broke some parts of the PyTorch integration.

On the other hand, this also allowed us to deal with further issues that contain breaking changes but needed to be done eventually. All of this should not cause you too much trouble, but still to make the transition as easy as possible there is a migration guide, which lists all breaking changes.

Added

  • General

    • Python requirements raised to Python 3.9 since Python 3.8 has reached end of life on 2024-10-07.

    • Dropped torchtext as an integration dependency. For individual use cases it can of course still be used.

    • Added environment variables SMALL_TEXT_PROGRESS_BARS and SMALL_TEXT_OFFLINE to control the default behavior for progress bars and model downloading.

  • PoolBasedActiveLearner:

    • initialize_data() has been replaced by initialize() which can now also be used to provide an initial model in cold start scenarios. (#10)

  • Datasets

    • Validation has been added to prevent TextDataset objects from containing None items (instead of str) either during initialization or when setting the x property. (#73)

  • Classification:

    • All PyTorch-classifiers (KimCNN, TransformerBasedClassification, SetFitClassification) now support torch.compile() which can be enabled on demand. (Requires PyTorch >= 2.0.0).

    • All PyTorch-classifiers (KimCNN, TransformerBasedClassification, SetFitClassification) now support Automatic Mixed Precision.

    • SetFitClassification.__init__() now has a verbosity parameter (similar to TransformerBasedClassification) through which you can control the progress bar output of SetFitClassification.fit().

    • TransformerBasedClassification:

      • Removed unnecessary token_type_ids keyword argument in model call.

      • Additional keyword args for config, tokenizer, and model can now be configured.

    • SetFitClassification:

      • Additional keyword args for trainer and model are now attached to SetFitModelArguments instead of SetFitClassification.

  • Embeddings:

    • Prevented unnecessary gradient computations for some embedding types and unified code structure.

  • Pytorch:

    • Added an inference_mode() context manager that applies torch.inference_mode or torch.no_grad for older Pytorch versions.

  • Query Strategies:

  • Vector Index Functionality:

    • A new vector index API provides implementations over a unified interface to use different implementations for k-nearest neighbor search.

    • Existing strategies that used a hard-coded vector search (ContrastiveActiveLearning, SEALS, AnchorSubsampling) have been adapted and can now be used with different vector index implementations.

Fixed

  • Fixed a bug where the clone() operation wrapped the labels, which then raised an error. This affected the single-label scenario for PytorchTextClassificationDataset and TransformersDataset. (#35)

  • Fixed a bug where the batching in greedy_coreset() and lightweight_coreset() resulted in incorrect batch sizes. (#50)

  • Fixed a bug where lightweight_coreset() failed when computing the norm of the elementwise mean vector.

Changed

  • Minimum required Python version was raised to 3.9.

  • General

    • Moved split_data() method from small_text.data.datasets to small_text.data.splits.

  • Dependencies

    • Raised setfit version to 1.1.0.

  • Classification:

    • The initialize() methods of all PyTorch-classifiers (KimCNN, TransformerBasedClassification, SetFitClassification) are now more unified. (#57)

    • KimCNNClassifier / TransformerBasedClassification: model selection is now disabled by default. Also, it no longer saves models when disabled, thereby greatly reducing the runtime.

  • Utils

    • init_kmeans_plusplus_safe() now supports weighted kmeans++ initialization for scikit-learn>=1.3.0.

  • Documentation:

    • Extended the section on initializing the active learning process.

Removed

  • Deprecated functionality

    • Removed default_tensor_type() method.

    • Removed small_text.utils.labels.get_flattened_unique_labels().

    • Removed small_text.integrations.pytorch.utils.labels.get_flattened_unique_labels().

    • Classification

      • Removed early stopping legacy arguments in __init__() for KimCNN and TransformerBasedClassification. (Use fit() keyword arguments instead.)

      • Removed model selection legacy argument in TransformerBasedClassification.__init__().

  • The explicit installation instruction for conda was removed, but the small-text conda-forge package will remain.

  • Removed gensim dependency in the PyTorch multiclass example.

Version 1.4.1 - 2024-08-18

Fixed

  • Fixed an out of bounds error that occurred when DiscriminativeActiveLearning queries all remaining unlabeled data.

  • Fixed typos/wording in PoolBasedActiveLearner docstrings.

  • Pinned SetFit version in notebook example. (#64)

  • Fixed an out of bounds error that could occur in SetFitClassification for both 32bit systems and Windows. (#66)

  • Fixed errors in notebook examples that occurred with more recent seaborn / matplotlib versions.

Changed

  • Documentation: added links to bibliography. (#65)

Contributors

@pdhall99

Version 1.4.0 - 2024-06-09

Added

Fixed

  • Changed the way how the seed is controlled in SetFitClassification since the seed was fixed unless explicitly set via the respective trainer keyword argument.

Changed

Version 1.3.3 - 2023-12-29

Changed

  • An errata section was added to the documentation.

Fixed

  • Fixed a deviation from the paper, where DeltaFScore also took into account the agreement in predictions of the negative label. (#51)

  • Fixed a bug in KappaAverage that affected the stopping behavior. (#52)

Contributors

@zakih2, @vmanc

Version 1.3.2 - 2023-08-19

Fixed

  • Fixed a bug in TransformerBasedClassification, where validations_per_epoch>=2 left the model in eval mode. (#40)

Version 1.3.1 - 2023-07-22

Fixed

  • Fixed a bug where parameter groups were omitted when using TransformerBasedClassification‘s layer-specific fine-tuning functionality. (#36, #38)

  • Fixed a bug where class weighting resulted in nan values. (#39)

Contributors

@JP-SystemsX

Version 1.3.0 - 2023-02-21

Added

Fixed

  • Fixed broken link in README.md.

  • Fixed typo in README.md. (#26)

Changed

  • The ClassificationChange stopping criterion now supports multi-label classification.

  • Documentation:

    • Updated the active learning setup figure.

    • The documentation of integrations has been reorganized.

Contributors

@rmitsch

Version 1.2.0 - 2023-02-04

Added

  • Added new classifier: SetFitClassification which wraps huggingface/setfit.

  • Active Learner:

    • PoolBasedActiveLearner now handles keyword arguments passed to the classifier’s fit() during the update() step.

  • Query Strategies:

  • Notebook Examples:

    • Revised both existing notebook examples.

    • Added a notebook example for active learning with SetFit classifiers.

    • Added a notebook example for cold start initialization with SetFit classifiers.

  • Documentation:

    • A showcase section has been added to the documentation.

Fixed

  • Distances in lightweight_coreset were not correctly projected onto the [0, 1] interval (but ranking was unaffected).

Changed

Version 1.1.1 - 2022-10-14

Fixed

  • Model selection raised an error in cases where no model was available for selection (#21).

Version 1.1.0 - 2022-10-01

Added

  • General:

    • Small-Text package is now available via conda-forge.

    • Imports have been reorganized. You can import all public classes and methods from the top-level package (small_text):

      from small_text import PoolBasedActiveLearner

  • Classification:

    • All classifiers now support weighting of training samples.

    • Early stopping has been reworked, improved, and documented (#18).

    • Model selection has been reworked and documented.

    • [!] KimCNNClassifier.__init()__: The default value of the (now deprecated) keyword argument early_stopping_acc has been changed from 0.98 to -1 in order to match TransformerBasedClassification.

    • [!] Removed weight renormalization after gradient clipping.

  • Datasets:

    • The target_labels keyword argument in __init()__ will now raise a warning if not passed.

    • Added from_arrays() to SklearnDataset, PytorchTextClassificationDataset, and TransformersDataset to construct datasets more conveniently.

  • Query Strategies:

  • Stopping Criteria:

Deprecated

  • small_text.integrations.pytorch.utils.misc.default_tensor_type() is deprecated without replacement (#2).

  • TransformerBasedClassification and KimCNNClassifier: The keyword arguments for early stopping (early_stopping / early_stopping_no_improvement, early_stopping_acc) that are passed to __init__() are now deprecated. Use the early_stopping keyword argument in the fit() method instead (#18).

Fixed

  • Classification:

    • KimCNNClassifier.fit() and TransformerBasedClassification.fit() now correctly process the scheduler keyword argument (#16).

Removed

  • Removed the strict check that every target label has to occur in the training data. (This is intended for multi-label settings with many labels; apart from that it is still recommended to make sure that all labels occur.)

Version 1.0.1 - 2022-09-12

Minor bug fix release.

Fixed

Links to notebooks and code examples will now always point to the latest release instead of the latest main branch.

Version 1.0.0 - 2022-06-14

First stable release.

Changed

  • Datasets:

    • SklearnDataset now checks if the dimensions of the features and labels match.

  • Query Strategies:

  • Documentation:

    • The documentation is now available in full width.

  • Repository:

    • Versions in this can now be referenced using the respective Zenodo DOI.

[1.0.0b4] - 2022-05-04

Added

  • General:

    • We now have a concept for optional dependencies which allows components to rely on soft dependencies, i.e. python dependencies which can be installed on demand (and only when certain functionality is needed).

  • Datasets:

    • The Dataset interface now has a clone() method that creates an identical copy of the respective dataset.

  • Query Strategies:

Changed

  • Datasets:

    • Separated the previous DatasetView implementation into interface (DatasetView) and implementation (SklearnDatasetView).

    • Added clone() method which creates an identical copy of the dataset.

  • Query Strategies:

    • EmbeddingBasedQueryStrategy now only embeds instances that are either in the label or in the unlabeled pool (and no longer the entire dataset).

  • Code examples:

    • Code structure was unified.

    • Number of iterations can now be passed via an cli argument.

  • small_text.integrations.pytorch.utils.data:

    • Method get_class_weights() now scales the resulting multi-class weights so that the smallest class weight is equal to 1.0.

[1.0.0b3] - 2022-03-06

Added

Changed

  • Cleaned up and unified argument naming: The naming of variables related to datasets and indices has been improved and unified. The naming of datasets had been inconsistent, and the previous x_ notation for indices was a relict of earlier versions of this library and did not reflect the underlying object anymore.

    • PoolBasedActiveLearner:

      • attribute x_indices_labeled was renamed to indices_labeled

      • attribute x_indices_ignored was unified to indices_ignored

      • attribute queried_indices was unified to indices_queried

      • attribute _x_index_to_position was named to _index_to_position

      • arguments x_indices_initial, x_indices_ignored, and x_indices_validation were renamed to indices_initial, indices_ignored, and indices_validation. This affects most methods of the PoolBasedActiveLearner.

    • QueryStrategy

      • old: query(self, clf, x, x_indices_unlabeled, x_indices_labeled, y, n=10)

      • new: query(self, clf, dataset, indices_unlabeled, indices_labeled, y, n=10)

    • StoppingCriterion

      • old: stop(self, active_learner=None, predictions=None, proba=None, x_indices_stopping=None)

      • new: stop(self, active_learner=None, predictions=None, proba=None, indices_stopping=None)

  • Renamed environment variable which sets the small-text temp folder from ALL_TMP to SMALL_TEXT_TEMP

[1.0.0b2] - 2022-02-22

Bugfix release.

Fixed

  • Fix links to the documentation in README.md and notebooks.

[1.0.0b1] - 2022-02-22

First beta release with multi-label functionality and stopping criteria.

Added

  • Added a changelog.

  • All provided classifiers are now capable of multi-label classification.

Changed

  • Documentation has been overhauled considerably.

  • PoolBasedActiveLearner: Renamed incremental_training kwarg to reuse_model.

  • SklearnClassifier: Changed __init__(clf) to __init__(model, num_classes, multi_Label=False)

  • SklearnClassifierFactory: __init__(clf_template, kwargs={}) to __init__(base_estimator, num_classes, kwargs={}).

  • Refactored KimCNNClassifier and TransformerBasedClassification.

Removed

  • Removed device kwarg from PytorchDataset.__init__(), PytorchTextClassificationDataset.__init__() and TransformersDataset.__init__().