Potential roadmap items 20162017+:

Predictive models

• Novelty detection using 1-class SVM  MADLIB-990
• Mixed effects modeling  MADLIB-987
• Factorization machines
• k-nearest neighbors (kNN)  MADLIB-927
• Geographically Weighted Regression (GWR) MADLIB-1023
• MCMC Probit and Logit regression
• Gaussian Mixture Model using Expectation Maximization (EM) algorithm MADLIB-410
• Multi-layer Perceptron
• Hierarchical clustering
• Neural nets and deep learning

Graph

• Single source shortest path  MADLIB-992
• Standard traversal
• depth first search
• breadth first search
topological sortAll pairs shortest path
• One mode projection (converting a bi-partitite graph of user-item graph to user-user or item-item graph)
• Connected components
• Page rank
• Hierarchical graph cut
• Graph cut
• Centrality measures like betweenness and closeness
• Triangle countingBetween-ness centrality
• Minimum spanning tree
• Graph diameter

Utilities

• Path functions (phase 2)  MADLIB-977
• Prediction metrics  MADLIB-907
• Sessionization  MADLIB-909
• Pivoting  MADLIB-908
• Anonymization  MADLIB-911
• URI tools  MADLIB-910
• Stratified sampling  MADLIB-986

...

• Expand coverage for PivotalR
• Expand coverage for PMML export??? (or perhaps switch to PFA???)
• Interface improvement and consistency for 2.0 release (does not need to be backward compatible)
• Implement an interface using named parameters
• Python API

...

• Work around PostgreSQL 1 GB field size limit  MADLIB-991
• GPU support
• Improve decision tree and random forest performance for run-time and memory usage

Platform

• Support for PostgreSQL 9.5 and 9.6  MADLIB-944
• Tensorflow support, or another deep learning framework