Features the work done by @Hopding (Andrew Dillon) to include his new SDRClassifier integrated into HTM.Java's Network API. This work allows the specification of not only which classifier to use but to be able to specify multiple fields to infer upon, and which classifier to use for each! (A different one can be specified for each "prediction field").
New Sync Release with build version incremented one extra due to administrative error in build file.
All is good with the world now...
HTM.Java Undergoes Total Vetting. 100% Compatibility Testing as of commit: Aug. 15th sha bba0c63dcc386cac216ab726366e6e32af98967
- NAB Scoring reaches compatibility with NuPIC
- Anomaly Detection is Fixed and 100% Compatible
- Added SDRClassifier
- Other significant fixes and optimizations (see the change log)
This is a quick release with minor enhancements for our partners at flink-htm. This release contains code which eases serialization when used within the Flink / Kryo environment.
- Also, turned off Standard Out by default during Gradle build
- New Persistence API
- New Docker Reference Build
- Convenience changes to build.gradle
- Many more... (see CHANGELOG.md for details)
This release contains fixes contributed by:
@antidata - Coordinate Encoder & Coordinate Encoder auto configuration fixes
@chelu - Sparse Matrix low memory implementations and optimizations
@cogmission - Coordinate Encoder Sensor fixes & Network implementation hot fix
Please see the Change Log for details...
While the API is not stable (meaning we may change it significantly at the slightest whim), the dependability is pretty stable and fit for non-critical production use. Please keep in mind that local changes may be necessary when upgrading to a newer pre-release version.