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 interchangable 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 Parallel interface for multiprocessing
• Added AdaMax neural network Optimizer
• 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 derivitive 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