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
• Single layer neural net estimators now use snapshotting
• 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 Placeholder 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 Guassian Naive Bayes
• Naive Bayes classifiers are now Online Estimators
• Added tolerance to Robust Z Score detector
• Added Concentration clustering metric (Calinski Harabasz)

0.0.2-alpha

Core neural net update, anomaly detection, and much more

0.0.1-alpha

Implemented head method on Supervised and Unsupervised dataset.