Goal / Scope
Final 1.9.1 release with new functionality and bug fixes.
Release Artifacts
Source code, rpm and dmg binaries are posted.
Date of release: September 19, 2016
JIRAs associated with the 1.9.1 release
Release Notes
MADlib v1.9.1
Release Date: 2016-SEP-19
New features:
* New function: One class SVM (MADLIB-990)
- Added a one-class SVM that classifies new data as similar or different to
the training set.
- This method is an unsupervised method that builds a decision boundary
between the data and origin in kernel space and can be used as a novelty detector.
* SVM: Added functionality to assign weights to each class, simplifying classification
of unbalanced data. (MADLIB-998)
* New function: Prediction metrics (MADLIB-907)
- Added a collection of summary statistics to gauge model accuracy based on
predicted values vs. ground-truth values.
* New function: Sessionization (MADLIB-909)
- Added a sessionize function to perform session reconstruction on a data
set, so that it can be prepared for input into other algorithms such as
path functions or predictive analytics algorithms.
* New function: Pivot (MADLIB-908)
- Added a function to that can do basic OLAP type operations on data stored
in one table and output the summarized data to a second table.
* Path: Major performance improvement (MADLIB-984)
* Path: Add support for overlapping patterns (MADLIB-995)
* Build: Add support for PG 9.5 and 9.6. Binaries avail for 9.5, build from source for 9.6 (MADLIB-944)
Bug fixes:
- Random Forest: Fix filtered feature related bug (MADLIB-928)
- Elastic Net: Skip arrays with NULL values in train (MADLIB-978)
- Matrix: Fix starting index in extract functions (MADLIB-1006)
- Path: Allow multiple expressions in partition expression (MADLIB-1003)
- DT: Fix bin computation for boolean features (MADLIB-1011)