NuPIC v0.5.6 Release Notes

    • 🚀 Since manylinux nupic.bindings wheel 0.4.10 has now been released to PyPi, we no longer need to install nupic.bindings from S3.
    • 🛠 fix logic in _getColumnNeighborhood
    • 🛠 Bugfix in flatIdx reuse after a segment is destroyed
    • 🔄 Change private _burstColumn class method signature to accept a cellsForColumn argument in lieu of a cellsPerColumn argument. Move the calculation that otherwise depends on cellsPerColumn into the instance method.
    • 👍 TM: Support extensibility by using traditional methods
    • ⚡️ Update expected error for topology changes
    • ⚡️ Update expected hotgym result for topology changes
    • ➕ Adds RELEASE.md with documentation for releasing NuPIC.
    • Match nupic.core's SP neighborhood ordering.
    • ⚡️ Update inhibition comments and docstrings.
    • 🚀 Introduce mechanism by which already-installed pre-release versions of nupic.bindings are ignored during installation
    • 👍 Assign self.connections value from self.connectionsFactory() rather than direct usage of Connections constructor. Allows better extensibility should the user want to change some aspect of the creation of the connections instance in a subclass
    • ✂ Removed obsolete directory src/nupic/bindings/
    • ✂ Remove the notion of "destroyed" Segments / Synapses
    • Enable proper subclassing by converting staticmethods that referenced TemporalMemory to classmethods that reference their class.
    • 🛠 Fixup TemporalMemory.write() to handle columnDimensions as tuples.
    • 🎉 Initialize columnDimensions as a tuple in test to reflect common convention. This forces the TemporalMemoryTest.testWriteRead test to fail in its current state.
    • Store "numActivePotentialSynapses". No more "SegmentOverlap".
    • ➕ Add a lot more scenarios to the TM perf benchmark
    • 🚚 Moved audiostream example to htm-community
    • Safer "addToWinners" value. Play nicely with surgical boosting.
    • 🛠 Bugfix: With no stimulus threshold, still break ties when overlaps=0
    • Clean up trailing whitespace and tabs
    • Properly apply the stimulus threshold
    • ➕ Add test for new "learn on predicted segments" behavior
    • Split compute into activateCells and activateDendrites
    • Grow synapses in predicted columns, not just bursting columns
    • ✂ Removed bundled get-pip.py and instead fetch version copy from S3
    • ✂ Removed .nupic_modules and now rely on versioned release of nupic.bindings on PyPI
    • ⚡️ averagingWindow size updated to improve HTM scores for RES-296
    • ⚡️ Build system updates for Bamboo (Linux), Travis (OS X), and AppVeyor (Windows)
    • ➕ Added nyc taxi example for anomaly detection