- 2.3.3 - Optimize Adam and AdaMax Optimizers - 2.3.2 - Update PHP Stemmer to version 3 - 2.3.1 - Fix PSR-3 log version compatibility issue - Check for correct version of RBX format - 2.3.0 - Added BM25 Transformer - Add `dropFeature()` method to the dataset object API - Add neural network architecture visualization via GraphViz - 2.2.2 - Fix Grid Search best model selection - 2.2.1 - Fix Extra Tree divide by zero when split finding - 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 - 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 - 2.0.2 - 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 warning instead of info - Added `header()` method to CSV and SQL Table Extractors - `Argmax()` now throws exception when undefined - MLP Learners recover from numerical instability with snapshot - Rename Gzip serializer to Gzip Native - Change RBX serializer constructor argument from base to level - Rename Writeable extractor interface to Exporter - 1.3.4 - Fix Decision Tree max height terminating condition - 1.3.3 - Forego unnecessary logistic computation in Logit Boost - 1.3.2 - Optimize Binary output layer - 1.3.1 - Update to Ok Bloomer 1.0 stable - 1.3.0 - Switch back to original fork of Tensor - Added `maxBins` hyper-parameter to CART-based learners - Added stream Deduplicator extractor - Added the SiLU activation function - Added Swish activation layer - 1.2.4 - Refactor neural network parameter updates - Allow set null logger - 1.2.3 - Fix Multiclass layer cross entropy gradient optimization - 1.2.2 - Allow empty dataset objects in `stack()` - 1.2.1 - Refactor stratified methods on Labeled dataset - Narrower typehints - 1.2.0 - Added Logit Boost classifier - Interval Discretizer variable or equi-depth binning - Text Normalizers now lower or upper case - 1.1.3 - Min Max Normalizer compensate for 0 variance features - 1.1.2 - Improved random floating point number precision - Deduplicate Preset seeder centroids - Fix Gradient Boost learning rate upper bound - Fix Loda histogram edge alignment - 1.1.1 - Fix Gradient Boost subsampling and importance scores - 1.1.0 - Update to Scienide Tensor 3.0 - Added Nesterov's lookahead to Momentum Optimizer - Added Reversible transformer interface - MaxAbs, Z Score, and Robust scalers are now Reversible - Min Max Normalizer now implements Reversible - TF-IDF Transformer is now Reversible - Added Preset cluster seeder - Added Concatenator extractor - 1.0.3 - Do not remove `groups` property from symbol table - 1.0.2 - Fix KNN and Hot Deck imputer reset donor samples - 1.0.1 - Fix AdaMax optimizer when tensor extension loaded - Prevent certain specification false negatives - Add extension minimum version specification - 1.0.0 - No changes - 1.0.0-rc1 - Added Token Hashing Vectorizer transformer - Added Word Stemmer tokenizer from Extras - Remove HTML Stripper and Whitespace Remover transformers - Rename steps() method to losses() - Steps() now returns iterable progress table w/ header - Remove rules() method on CART - Removed results() and best() methods from Grid Search - Change string representation of NAN to match PHP - Added extra whitespace pattern to Regex Filter - 1.0.0-beta2 - Interval Discretizer now uses variable width histograms - Added TF-IDF sublinear TF scaling and document length normalization - Dataset filterByColumn() is now filter() - Added Lambda Function transformer from Extras - Rename Dataset column methods to feature - Added Dataset general sort() using callback - Confusion Matrix classes no longer selectable - Remove Recursive Feature Eliminator transformer - Metric range() now returns a Tuple object - 1.0.0-beta1 - Added variance smoothing to Gaussian NB, Mixture, and MLE - Added MAD smoothing to Robust Z Score - Added Writable extractor interface - NDJSON and CSV extractors are now Writable - Added SQL Table dataset extractor - Changed Word Count Vectorizer DF constraints to proportions - Change order of Naive Bayes hyper-parameters - Persisters use RBX serializer by default - Removed previously deprecated portions of the API - Removed Embedder interface and namespace - Change Robust Z Score alpha parameter name to beta - Hold Out validator does not randomize by default - Move Redis DB persister to extras package - Remove Loda estimate bins static method - Change Grid Search base estimator param name to class - Remove Dataset cast to string preview - Add Error Analysis error standard deviation and drop midrange - Naive Bayes Laplace smoothing no longer effects priors - Nearest Neighbors distance weighting off by default - Promoted the Other namespace - Moved Flysystem persister to the Extras package - Change order of Loda hyper-parameters - Persistent Model now accepts an optional serializer - Persisters no longer interact directly with Persistables - Remove Wrapper interface - RBX serializer now accepts base Gzip parameter - Gzip serializer no longer accepts base serializer - Changed Gzip default compression level from 1 to 6 - Changed RBX default compression level from 9 to 6 - Do not persist training progress information - Change underscores in Report property names to spaces - Add saveTo() method to Encoding object - Add Dataset exportTo() method - Pipeline and Committee Machine are no longer Verbose - Remove K Best feature selector (special case of RFE) - Changed Error Analysis metrics - Remove threat score from Multiclass Breakdown - Rename Labels Are Missing exception - Feature importances are no longer normalized - Optimized CART binary categorical node splitting - Interval Discretizer outputs numeric string categories - Renamed Random Hot Deck Imputer - Changed order of decision tree hyper-parameters - 0.4.1 - Optimized CART node splitting for low variance continuous features - Fixed RBX serializer string representation - Prevent overwrites when instantiating Unlabeled from iterator - 0.4.0 - Added Truncated SVD transformer - Added Rubix Object File (RBX) format serializer - Added class revision() method to the Persistable interface - Added custom class revision mismatch exception - Add Boolean Converter transformer - Deprecated Igbinary serializer and move to Extras package - Deprecate explainedVar() and noiseVar() methods on PCA and LDA - Added missing extension specification and exception - 0.3.2 - Fix t-SNE momentum gain bus error when using Tensor extension - Optimize t-SNE matrix instantiation - Refactor single sample inference methods - Update the docs site - 0.3.1 - Fix CART feature importances purity increase overflow - 0.3.0 - Added K Best feature selector - Added Flysystem 2.0 Persister - Stateful and Elastic Transformers are now Persistable - Added Gzip serializer for Persistable objects - Added Sentence tokenizer - Library now throws Rubix\ML namespaced exceptions - Added Scoring interface for estimators that score samples - Deprecated the Ranking interface - Add generic Trainable interface - Decision Trees are now iterable - Added K-Skip-N-Gram tokenizer and deprecated Skip Gram - Single sample inference methods are now marked internal - Deprecated Variance Threshold Filter - 0.2.4 - Categorized and annotated internal API - Fix context of preprocess() and combinations() methods - Added version constants - 0.2.3 - Now compatible with PHP 8 GD Image types - Dataset cast sample to array upon validation - 0.2.2 - Optimized CART quantile-based node splitting - Fixed CART and Extra Tree min purity increase post pruning - Fix ITree infinite loop splitting same samples - 0.2.1 - Optimized Stop Word Filter - Allow list of empty regex patterns in Regex Filter - Handle missing class definitions in Native and Igbinary - Fixed infinite loop in Ball Tree & KD Tree grow method - 0.2.0 - Add Recursive Feature Eliminator feature selector - Can now disable holdout validation in MLP learners - TF-IDF Transformer additive Laplace smoothing now variable - Added instability detection to gradient-based learners - Gradient Boost validation set holdout can now be 0 - Specifications now extend base class - Rename Dataset validate argument to verify - Ball Tree Cluster nodes are now called Cliques - ITree cells are now called Depth nodes - Added Dataset join() method and deprecated augment() - Added score() method to Ranking API and deprecated rank() - Renamed Radius Neighbors anomalyClass to outlierClass - HTML Stripper can now allow user-specified tags - Sparse Random Projector now has variable sparsity - Deprecated Dense Random Projector transformer - 0.1.6 - Fix KNN Imputer spatial tree dependency injection - 0.1.5 - Compensate for zero vectors in Cosine kernel - Fixed KMC2 random threshold calculation - Fix Naive Bayes divide by zero when smoothing is 0 - 0.1.4 - Optimized Cosine distance for sparse vectors - 0.1.3 - Optimized Cosine distance kernel - Optimized (NaN) Safe Euclidean distance kernel - Fixed markedness calculation in Multiclass Breakdown - Prevent infinite loop during spatial tree path finding - 0.1.2 - Fixed Grid Search best hyper-parameters method - Fixed K Means average loss calculation - Fixed bootstrap estimators tiny bootstrap sets - 0.1.1 - Fixed Image Resizer placeholder image - Fixed Filesystem no write permissions on instantiation - Nicer Stringable object string representations - Do not terminate empty Spatial tree leaf nodes - Additional Filesystem persister checks - Nicer Dataset object validation error messages - 0.1.0 - CV Report Generators now return Report objects - Dataset describe methods now return Report objects - Allow hyphens and apostrophes in Word Tokenizer - Dataset conversion methods now return an Encoding object - Encodings are now writeable to disk - Allow classes to be selected for Confusion Matrix - Fixed divide by zero in Multiclass Breakdown report - Changed Random Projector minDimensions default max distortion - Fixed Naive Bayes user-defined class prior probabilities - Internal CV Learners now check for sufficient hold out data - Fixed randomize empty dataset object - Removed setPersister method from Persistent Model - Added Dataset Has Dimensionality Specification - Changed name of Tree max depth parameter to max height - Fixed F Beta division by zero - Dataset toCSV and toNDJSON accept optional header - Nicer Verbose Learner logger output - Screen Logger uses empty channel name by default - 0.1.0-rc5 - Improved logging for Verbose Learners - Added max document frequency to Word Count Vectorizer - Whitespace Trimmer is now a separate Transformer - Text Normalizers no longer remove extra whitespace - Added extra characters pattern to Regex Filter class constants - Moved Lambda Function transformer to Extras package - GaussianNB new class labels during partial train - Decision Tree print ruleset now accepts a header - Fixed Variance Threshold Filter drop categorical by default - Removed AdaBoost return learned sample weights - 0.1.0-rc4 - Added Multibyte Text Normalizer transformer - V Measure now has adjustable beta parameter - Persistent Model is no longer Verbose - Stop Word Filter now handles unicode characters - 0.1.0-rc3 - Embedders now adopt the Transformer API - Added RanksFeatures interface - Logistic Regression and Adaline now implement RanksFeatures - Ridge now implements the RanksFeatures interface - Added L2 regularization to Dense hidden layers - Neural Network L2 regularization now optional - Added MLP numerical instability checks - Optimized Ball Tree nearest neighbors search - Pipeline is now more verbose - Renamed Dataset partition method to partitionByColumn - Decreased default neural net learner batch size to 128 - Increased default K Means batch size to 128 - Renamed Dataset types method to featureTypes - Efficient serialization of Word Count Vectorizer - Decoupled Persistable interface from Learner - Moved Gower Distance kernel to Extras package - Moved SiLU activation function to Extras package - Removed array_first and array_last from global functions - Abstracted deferred Backend computations into Tasks - Removed unused BST interface - 0.1.0-rc2 - Persistent Model now implements Verbose interface - Tuned CART continuous feature quantile-based split finding - N-gram and SkipGram use configurable base word tokenizer - Moved Alpha Dropout hidden layer to Extras package - Added Dataset merge and augment methods - Removed Dataset prepend and append methods - Lambda Function transformer now takes any callable - Text Normalizer trim extra whitespace not optional - Mean Shift minimum seeds now set at 20 - Standardized K Means inertial loss over batch count - Added set persister method to Persistent Model - Removed range() from neural network Cost Function interface - Increased default neural net learner batch size to 200 - 0.1.0-rc1 - Random Forest now handles imbalanced datasets - Added early stopping window to AdaBoost - Gaussian MLE now has automatic and adaptive threshold - Loda now has automatic and adaptive threshold - Variance Threshold Filter now selects top k features - Added params method to Estimator and Embedder interface - t-SNE now compatible with categorical distance kernels - Grid Search implements the Wrapper interface - Grid Search memorizes all results from last search - Dataset fromIterator method accepts any iterable - Column Picker throws exception if column not found - Better hyper-parameter stringification - Improved Dataset exception messages - RMSE now default validation Metric for Regressors - Added balanced accuracy and threat score to Multi-class report - Pipeline and Persistent Model now implement Ranking - Changed percentile to quantile in Stats helper - Renamed Residual Analysis report to Error Analysis - Changed namespace of specification objects - 0.0.19-beta - Added SiLU self-stabilizing neural network activation function - Dense hidden layers now have optional bias parameter - KNN-based imputers accelerated by spatial tree - Changed the default anomaly class for Radius Neighbors - Removed additional methods from guessing Strategies - Numeric String Converter now uses fixed NaN placeholder - Missing Data Imputer now passes through other data types - Changed order of Missing Data Imputer params - Renamed high-level resource type to image type - Added comb (n choose k) to global functions - Image Vectorizer now has grayscale option - Clusterers and Anomaly Detectors return integer predictions - Ball Tree now compatible with categorical distance kernels - Parallel Learners using Amp Backend are now persistable - Changed order of Radius Neighbors hyper-parameters - 0.0.18-beta - Now requires PHP 7.2 and above - Added phpbench performance benchmarks - Added JSON, NDJSON, CSV, and Column Picker Extractors - Changed the way fromIterator method works on Dataset object - Added Hyperplane dataset generator - Changed the way noise is applied to Circle, Half Moon, etc. - Changed name of Multilayer Perceptron classifier - Deferred computations are now callable - Removed range() from the activation function interface - Added label type validation for supervised learners - Added toArray, toJson, toCsv, toNdjson methods to Dataset API - Can now preview a Dataset object in console by echoing it - Changed Labeled dataset objects iteration and array access - Removed zip and unzip methods on Labeled dataset - Added describe by label method to Labeled dataset - Changed the way fromIterator works on Dataset - Added Regex Filter transformer - Changed name of Igbinary serializer - Changed dataset and label description - 0.0.17-beta - Added Tensor extension compatibility - Migrated to new Tensor library namespace - Anomaly detector predictions now categorical - Clusterers now predict categorical cluster labels - Added extracting data section to docs - Added code metrics - Added training and inference sections to the docs - Decision tree rules method now outputs a string - Added drop row and column methods to dataset interface - Dataset row() method is now sample() - 0.0.16-beta - Radius Neighbors allows user-definable anomaly class - Added KNN Imputer - Added Random Hot Deck Imputer - Missing Data Imputer now handles NaNs by default - Added NaN safe Euclidean distance kernel - Added Gower distance kernel - Added Hamming distance kernel - Dataset now requires homogeneous feature columns - KNN now compatible with categorical features - Added transform column method to dataset object - Added describe method to dataset object - Added describe labels method to Labeled dataset - Added deduplicate method to dataset object - Added unzip static factory for Labeled datasets from data table - Changed the order of t-SNE hyper-parameters - Added global transpose array helper function - Renamed label key to classes in Multiclass Breakdown report - Changed order of Gradient Boost and AdaBoost hyper-parameters - Changed order of Loda hyper-parameters - Added asString method to the Data Type helper class - Added check for NaN labels in Labeled dataset - Changed namespace of Data Type helper - Numeric String Converter now handles NaN strings - Added predict probabilities of a single sample method - Added rank single sample trait - 0.0.15-beta - Added Gaussian MLE anomaly detector - Added early stopping window to Gradient Descent-based Learners - Changed early stopping behavior of MLP-based estimators - Added predict single sample method to Learner interface - Changed method signature of random subset without replacement - Changed K Means default max iterations - Robust Z-Score now uses weighted combination of scores - Cross validators now stratify dataset automatically - Changed default k in K Fold validator - Changed order of Loda hyperparameters - Changed hyperparameter order of KNN-based learners - Added method to return categories from One Hot Encoder - Removed Lottery and Blurry Percentile guessing strategy - Added Percentile guessing strategy - Added shrinkage parameter to Wild Guess strategy - Added additional methods to random Strategies - Renamed Popularity Contest strategy to Prior - Datasets now inherit from abstract parent Dataset class - Removed Dataset interface - Neural net parameter update in Layer instead of Optimizer - Changed order of distance-based clusterer hyperparameters - Improved cluster radius estimation in Mean Shift - Naive Bayes now adaptive to new class labels - Changed order of neural network learner hyperparameters - Added safety switch to AdaBoost if weak learner worse than random - Added min change early stopping to AdaBoost - Added Patreon funding support - 0.0.14-beta - Added feature importances to Gradient Boost - Added progress monitoring to Gradient Boost w/ early stop - Added Spatial and Decision tree interface - Mean Shift compatible with Spatial trees - K-d Neighbors base spatial tree configurable - Radius Neighbors now uses base spatial tree - Local Outlier Factor interchangeable base search tree - DBSCAN now uses any Spatial tree for range searches - CART uses downsampling on continuous features - LOF and Isolation Forest contamination off by default - Embed method now returns an array instead of dataset - Fixed issue with Dataset partitioning - Renamed Coordinate node to Hypercube - KNN default k is now 5 instead of 3 - CART can now print a text representation of the decision rules - Removed Local Outlier Factor brute force version - Changed namespace of trees to Graph/Trees - CART impurity tolerances are now hardcoded - Changed order of CART hyperparameters - Added Extra Tree base implementation - Extra Tree splits are now unbiased - Extra Tree Classifier now minimizes entropy - Reduced the memory footprint of Binary Nodes - Gradient Boost shrinkage bounded between 0 and 1 - Added random subset without replacement to dataset API - Changed order of Gradient Boost hyperparameters - Changed order of MLP hyperparameters - Ranking interface is now a general interface - Changed default t-SNE minimum gradient - 0.0.13-beta - Added documentation site - Added Regression and Classification Loss interfaces - Robust Z-Score is now a Ranking anomaly detector - Loda now defaults to auto detect bin count - Removed tolerance param from Gradient Boost and AdaBoost - Screen logger timestamp format now configurable - Dropped Persistable contract between SVM-based learners - Random Forest feature importances now serial - Removed Robust Z-Score tolerance parameter - Added slice method to Dataset API - Loda now performs density estimation on the fly - Transform labels now returns self for method chaining - 0.0.12-beta - Added AdaMax neural network Optimizer - Added Parallel interface for multiprocessing - Added Backend processing interface - Added Amp parallel and Serial processing Backends - Random Forest uses parallel processing - Added CPU helper and core auto detection - Committee Machine is now a meta estimator - Committee Machine now Parallel and Verbose - Bootstrap Aggregator uses multiple processes - Grid Search now trains in parallel - K Fold, Leave P Out, and Monte Carlo validators now Parallel - Added momentum to Batch Norm moving averages - Custom Batch Norm and PReLU parameter initialization - Added custom bias initialization to Dense layer - Output layers now accept custom initializers - Added Constant neural network parameter initializer - Removed Exponential neural network Cost Function - Filesystem save history is now either on or off - Removed save history from Redis DB Persister - Removed Model Orchestra meta-estimator - Grid Search automatically retrains base estimator - Added neural net Parameter namespace and interface - Changed order of Loda hyperparameters - Replaced F1 Score with F Beta metric - Removed ISRU and Gaussian activation functions - Fixed SELU derivative computation - Changed adaptive optimizer default decay parameters - Changed default learning rate of Stochastic Optimizer - Added SMAPE (Symmetric MAPE) regression metric - Added MAPE to Residual Analysis report - Fixed MSLE computation in Residual Analysis report - Renamed RMSError Metric to RMSE - Embedders no longer implement Estimator interface - Added error statistics to Residual Analysis report - 0.0.11-beta - K Means now uses mini batch GD instead of SGD - K Means in now an Online learner - Added Adjusted Rand Index clustering metric - Added Seeder Interface - Added Random, K-MC2, and Plus Plus seeders - Accelerated Mean Shift with Ball Tree - Added radius estimation to Mean Shift - K Means and Mean Shift now implement Probabilistic - Gaussian Mixture now supports seeders - Changed order of K Means hyperparameters - Moved Ranking interface to anomaly detector namespace - N-gram Tokenizer now outputs ranges of word tokens - Changed default Fuzzy C Means hyper-parameters - Added spatial partitioning to Dataset API - Added Image Resizer transformer - Image Vectorizer no longer resizes images - Fixed adaptive optimizer bug upon binary unserialization - Removed Quartile Standardizer - Optimized Image Vectorizer using bitwise operations - Pipeline is now more verbose - 0.0.10-beta - Added Loda online anomaly detector - Added Radius Neighbors classifier and regressor - Added fast k-d LOF anomaly detector - Added base Ball Tree implementation - Added Ranking interface - Changed Manifold namespace to Embedders - Isolation Forest and LOF are now Ranking - K Means is now Verbose - Accelerated DBSCAN with Ball Tree - Added upper bound to contamination hyperparameter - Changed hyper-parameter order of Isolation Forest - Optimized Interval Discretizer transformer - K Means is no longer Online - Removed Sign function - Added Binary Tree interface - Added bin count heuristic to Loda - Changed order of k-d neighbors hyperparameters - Removed Hamming distance kernel - 0.0.9-beta - Added transform labels method to Labeled Dataset - Added Data Type helper - Pipeline and Persistent Model are now Probabilistic - Added stack method to dataset API - Changed merge method on dataset to append and prepend - Implemented specifications - Added data type compatibility for estimators - Added compatibility method to validation metrics - Added estimator compatibility to reports - Added trained method to learner API - Added fitted method to Stateful transformer API - Changed ordinal of integer encoded data types - Added Adaptive optimizer interface - Changed Transformer transform API - Removed prompt method from Persistent Model - Removed JsonSerializable from Dataset Interface - 0.0.8-alpha - Added Model Orchestra meta estimator - Added Stop Word Filter transformer - Added document frequency smoothing to TF-IDF Transformer - Added Uniform neural net weight initializer - Improved Gaussian Mixture numerical stability - Fixed missing probabilities in Classification Tree - Removed MetaEstimator interface - Added model Wrapper interface - AdaBoost is now probabilistic - Added Constant guessing strategy - Added N-Gram word tokenizer - Added Skip-Gram word tokenizer - Changed FCM and K Means default max epochs - Added zip method to Labeled dataset - Removed stop word filter from Word Count Vectorizer - Changed order of t-SNE hyper-parameters - Grid search now has automatic default Metric - Base k-D Tree now uses highest variance splits - Renamed Raw Pixel Encoder to Image Vectorizer - 0.0.7-alpha - Added Support Vector Machine classifier and regressor - Added One Class SVM anomaly detector - Added Verbose interface for logging - Added Linear Discriminant Analysis (LDA) transformer - Manifold learners are now considered Estimators - Transformers can now transform labels - Added Cyclic neural net Optimizer - Added k-d neighbors search with pruning - Added post pruning to CART estimators - Estimators with explicit loss functions are now Verbose - Grid Search: Added option to retrain best model on full dataset - Filesystem Persister now keeps backups of latest models - Added loading backup models to Persister API - Added PSR-3 compatible screen logger - Grid Search is now Verbose - t-SNE embedder is now Verbose - Added Serializer interface - Added Native and Binary serializers - Fixed Naive Bayes reset category counts during partial train - Pipeline and Persistent Model are now Verbose - Classification and Regression trees now Verbose - Random Forest can now return feature importances - Gradient Boost now accepts base and booster estimators - Blurry Median strategy is now Blurry Percentile - Added Mean strategy - Removed dataset save and load methods - Subsumed Extractor api into Transformer - Removed Concentration metric - Changed Metric and Report API - Added Text Normalizer transformer - Added weighted predictions to KNN estimators - Added HTML Stripper transformer - 0.0.6-alpha - Added Gradient Boost regressor - Added t-SNE embedder - AdaBoost now uses SAMME multiclass algorithm - Added Redis persister - Added Max Absolute Scaler - Added Principal Component Analysis transformer - Pipeline is now Online and has elastic option - Added Elastic interface for transformers - Z Scale Standardizer is now Elastic - Min Max Normalizer is now Elastic - TF-IDF Transformer is now Elastic - Added Huber Loss cost function - Added Swiss Roll generator - Moved Generators to the Datasets directory - Added Persister interface for Persistable objects - Added overwrite protection to Persistent Model meta estimator - Multiclass Breakdown report now breaks down user-defined classes - Renamed restore method to load on Datasets and Persisters - Random Forest now accepts a base estimator instance - CARTs now use max features heuristic by default - Added build/quick factory methods to Datasets - Added Interval Discretizer transformer - GaussianNB and Naive Bayes now accept class prior probabilities - Removed Image Patch Descriptor - Added Learner interface for trainable estimators - Added smart cluster initialization to K Means and Fuzzy C Means - Circle and Half Moon generators now generate Labeled datasets - Gaussian Mixture now uses K Means initialization - Removed Isolation Tree anomaly detector - 0.0.5-alpha - Added Gaussian Mixture clusterer - Added Batch Norm hidden layer - Added PReLU hidden layer - Added Relative Entropy cost function to nn - Added random weighted subset to datasets - Committee Machine classifier only and added expert influence - Added type method to Estimator API - Removed classifier, detector, clusterer, regressor interfaces - Added epsilon smoothing to Gaussian Naive Bayes - Added option to fit priors in Naive Bayes classifiers - Added Jaccard distance kernel - Fixed Hamming distance calculation - Added Alpha Dropout layer - Fixed divide by 0 in Cross Entropy cost function - Added scaling parameter to Exponential cost function - Added Image Patch Descriptor extractor - Added Texture Histogram descriptor - Added Average Color descriptor - Removed parameters from Dropout and Alpha Dropout layers - Added option to remove biases in Dense and Placeholder1D layers - Optimized Dataset objects - Optimized matrix and vector operations - Added grid params to Param helper - Added Gaussian RBF activation function - Renamed Quadratic cost function to Least Squares - Added option to stratify dataset in Hold Out and K Fold - Added Monte Carlo cross validator - Implemented noise as layer instead of activation function - Removed Identity activation function - Added Xavier 1 and 2 initializers - Added He initializer - Added Le Cun initializer - Added Normal (Gaussian) initializer - 0.0.4-alpha - Added Dropout hidden layer - Added K-d Neighbors classifier and regressor - Added Extra Tree Regressor - Added Adaline regressor - Added sorting by column to Dataset - Added sort by label to Labeled Dataset - Added appending and prepending to Dataset - Added Dataset Generators - Added Noisy ReLU activation function - Fixed bug in dataset stratified fold - Added stop word filter to Word Count Vectorizer - Added centering and scaling options for standardizers - Added min dimensionality estimation on random projectors - Added Gaussian Random Projector - Removed Ellipsoidal distance kernel - Added Thresholded ReLU activation function - Changed API of Raw Pixel Encoder - 0.0.3-alpha - Added Extra Tree classifier - Random Forest now supports Extra Trees - New Decision Tree implementation - Added Canberra distance kernel - Committee Machine is now a Meta Estimator Ensemble - Added Bootstrap Aggregator Meta Estimator Ensemble - Added Gaussian Naive Bayes - Naive Bayes classifiers are now Online learners - Added tolerance to Robust Z-Score detector - Added Concentration clustering metric (Calinski Harabasz) - 0.0.2-alpha - Added Anomaly Detection - New Neural Net implementation - Added static analysis - Added Travis CI configuration - 0.0.1-alpha