Goal / Scope
Final 1.10 release with new functionality and bug fixes.
Release Artifacts
Source code, rpm and dmg binaries are posted.
Date of release: Mar 10, 2017
JIRAs associated with the 1.10 release
Release Notes
MADlib v1.10
Release Date: 2017-MAR-10
New features:
* New module: Graph - Single Source Shortest Path (SSSP) (MADLIB-992)
- Calculate the shortest path from a given vertex to every vertex in the graph.
* New module: Encode categorical variables (MADLIB-1038)
- Completely new version for dummy/one-hot encoding of categorical variables with new name and different arguments.
- Previous version has been deprecated.
* New module (early stage): K-Nearest Neighbors (KNN) (MADLIB-927)
- Find the k nearest neighbors based on the squared_dist_norm2 metric.
* Elastic Net: Add grouping support (MADLIB-950)
- Elastic net train for both Gaussian and Binomial models, with FISTA
and IGD optimizations support grouping.
- Use active sets for FISTA, but active sets are used only after the
log-likelihood of all the groups becomes 0.
* Elastic Net: Add cross validation (MADLIB-996)
* PCA: Add grouping support (MADLIB-947)
* PCA: Removed column id restriction.
* Kmeans: State the value of cluster variance to MADlib.
* Kmeans: Support for array input. (MADLIB-1018)
* DT and RF: Verbose option for the dot output format. (MADLIB-1051)
* Association Rules: Add rule counts and limit itemset size feature (MADLIB-1044, MADLIB-1031)
* Boost library has been upgraded from 1.47 to 1.61
* Multiple improvements to the build system (madpack, cmake etc.) to support Semantic versioning and various versions of GPDB and HAWQ.
Bug fixes:
- Pivot: Adjust the warning level to remove redundant messages.
- RF: Fix the online help and examples.
- Utilities: Fix incorrect flag for distribution.
- Install check: Update date format and remove hardcoded schema names.
- Multiple user documentation improvements.