0.4.3

• Update to Flysystem 2.1 and above

2.2.0

• Added Image Rotator transformer
• Added One Vs Rest ensemble classifier
• Add variance and range to the Dataset describe() report
• Added Gower distance kernel
• Added types() method to Dataset
• Concatenator now accepts an iterator of iterators

2.1.1

• Do not consider unset properties when determining revision

2.1.0

Big thanks to @torchello and @DrDub for their huge contributions to this release!

• Added Probabilistic Metric interface
• Added Probabilistic and Top K Accuracy
• Added Brier Score Probabilistic Metric
• Export Decision Tree-based models in Graphviz "dot" format
• Added Graphviz helper class
• Graph subsystem memory and storage optimizations

Warning: This release contains changes to the Graph subsystem which breaks backward compatibility for all Decision tree-based learners that were saved with a previous version. Classification Tree, Extra Tree Classifiers, Random Forests, LogitBoost, Adaboost, Regression Tree, Extra Tree Regressor, and Gradient Boost are all affected.

Note: Moving forward, we will only release changes that break the backward compatibility of saved objects in a major release unless they are part of a bug fix. See https://docs.rubixml.com/2.0/model-persistence.html#caveats for an explanation as to why saved objects are not as straightforward to maintain backward compatibility as the API.

2.0.2

• Fix Decision Tree max height terminating condition

1.3.4

• Fix Decision Tree max height terminating condition

2.0.1

• Compensate for PHP 8.1 backward compatibility issues

2.0.0

• Gradient Boost now uses gradient-based subsampling
• Allow Token Hashing Vectorizer custom hash functions
• Gradient Boost base estimator no longer configurable
• Move dummy estimators to the Extras package
• Increase default MLP window from 3 to 5
• Decrease default Gradient Boost window from 10 to 5
• Rename alpha regularization parameter to L2 penalty
• Added RBX serializer class property type change detection
• Rename boosting estimators param to epochs
• Neural net-based learners can now train for 0 epochs
• Rename Labeled stratify() to stratifyByLabel()
• Added Sparse Cosine distance kernel
• Cosine distance now optimized for dense and sparse vectors
• Word Count Vectorizer now uses min count and max ratio DFs
• Numeric String Converter now handles NAN and INFs
• Numeric String Converter is now Reversible
• Removed Numeric String Converter NAN_PLACEHOLDER constant
• Added MurmurHash3 and FNV1a 32-bit hashing functions to Token Hashing Vectorizer
• Changed Token Hashing Vectorizer max dimensions to 2,147,483,647
• Increase SQL Table Extractor batch size from 100 to 256
• Ranks Features interface no longer extends Stringable
• Verbose Learners now log change in loss
• Numerical instability logged as a warning instead of info
• Added header() method to CSV and SQL Table Extractors
• Argmax() now throws an exception when undefined
• MLP Learners recover from numerical instability with a snapshot
• Rename Gzip serializer to Gzip Native
• Change RBX serializer constructor argument from base to level
• Rename Writeable extractor interface to Exporter

1.3.2

• Forego unnecessary logistic computation in Logit Boost

0.4.2

• Fix Missing Extension exception class filename