Skip to content

Latest commit

 

History

History
406 lines (338 loc) · 17.3 KB

CHANGELOG.md

File metadata and controls

406 lines (338 loc) · 17.3 KB
Raw

Changelog

0.5.2

  • Fixe to GCE to return the right number of scalars when altitude is missing.

0.5.1

  • Improves SDR classifier and tests
  • Modify the continuous online learning test
  • Add 3 tests on multiple item prediction
  • Fix test_pFormatArray
  • Implement SDR classifier in NuPIC
  • Make the 'arrayTypes' list more informative
  • Add getParameter/setParameter support for Bool and BoolArray
  • Improved anomaly params (from NAB)
  • Added minSparsity option
  • Get the encoder's outputWidth via parameter
  • Use nupic.core encoders from nupic via the Network API
  • Fix bugs and inconsistencies in the custom region demo
  • Adds BINDINGS_VERSION envvar to wheel filename (for iterative builds)

0.5.0

  • Removes references to FastTemporalMemory.
  • Lower TM epsilon threshold for compatibility.
  • Add documentation for the Monitor Mixins
  • Removed FastTemporalMemory from nupic
  • Update temporal memory compatibility test to use C++ TM.
  • Sort segments before iterating for compatibility with C++
  • Sort unpredictedActiveColumns before iterating for compatibility with C++

0.4.5

  • This release is just to sync with nupic.bindings 0.3.1.

0.4.4

  • Botched release (sorry!)

0.4.3

  • Updating to proper core sha

0.4.2

  • Using official release version of bindings for nupic release.

0.4.1

  • Manualy update of nupic.bindings dev version after botched release attempt

0.4.0

  • Updated hello_tm.py to use accessors
  • Updated TP_shim.py to use accessors Updated columnForCell and _validateCell in FastTemporalMemory to conform to their docstrings, which is needed for the change to TP_shim.py
  • Updated temporal memory monitor mixin to use accessors
  • Updated temporal_memory_test.py to use accessor methods.
  • Added accessors to temporal_memory.py
  • Change temporalImp to tm_py for both networks and add comment about it being a temporary value until C++ TM is implemented
  • Refactored to remove common code between network_checkpoint_test.py and temporal_memory_compatibility_test.py
  • Use named constants from nupic.data.fieldmeta in aggregator module instead of naked constants.
  • Fix AttributeError: 'TPShim' object has no attribute 'topDownCompute'
  • Support more parameters in TPShim
  • Serialize remaining fields in CLAModel using capnproto
  • Enforce pyproj==1.9.3 in requirements.txt
  • Use FastCLAClassifier read class method instead of instance method
  • Have CLAClassifierFactory.read take just the proto object
  • Add capnp serialization to CLAClassifierRegion
  • Add capnp serialization to SPRegion

0.3.6

  • Windows support
  • Serialization work
  • Moved SWIG out into nupic.core
  • Major build changes

0.3.5

  • Raise explicit exception if user passes non-str path
  • SP: simplify local inhibition
  • SP: adapt tests, sort winning columns output
  • SP: simplify active columns assignment
  • SP: simplify global inhibition
  • file Rename as hello_tm.py and modifications in comments

0.3.4

  • Added src/nupic/frameworks/opf/common_models/cluster_params.py and supporting files from numenta-apps htmengine. A separate numenta-apps PR will remove this code from htmengine.
  • fixes #2592
  • fix for #2265
  • fix for bug #2265
  • Fixup Dockerfile to install nupic.bindings, and other cleanup
  • Adding C++ compiler requirement to README.
  • Fix for test failure
  • Fixed stream definition reference error.
  • Reduce default reestimation period.
  • Remove greedy reestimation of distribution
  • Pointing README to proper bindings version.
  • Continuing work on 0.3.4.dev0.
  • removing a test that depends on nupic.vision
  • PCA_Node test: some fixes, WIP
  • formatting
  • test for PCANode region
  • remove Pillow from requirements.txt as it was used for vision only
  • fix merge mistake in csv file
  • move test from PCANode to nupic.vision unittest

0.3.3

  • Include additional file types in MANIFEST.in, consistent with setup.py
  • Pattern and Sequence machines using nupic::Random
  • Wrap sparse matrix implementations with cortical column-centric semantics as a way to abstract away the underlying implementation
  • Re-enable testHotgymRegression

0.3.2

  • Update to nupic.bindings version with fix for platform differences
  • Rename nupic directory to src/nupic
  • Updated S3 URL to nupic.bindings for Linux install
  • Fix paths for data files in an integration test
  • Fix issue with storing temporary file in wrong location in integration test

0.3.1

  • Specify nupic.bindings version to match commit sha (0.2).
  • Use logging.debug for emitting the message about not being able to import matplotlib; we log it at debug level to avoid polluting the logs of apps and services that don't care about plotting.
  • Add Dockerfile ready to perform swarming.
  • Removes PCANode
  • Updated Linux binary install instructions.

0.3.0

  • Updated comment about greedy stats refresh when likelihood > 0.99

0.2.12

  • Implemented unit tests for the new features in AnomalyLikelihood class.
  • Convert AnomalyLikelihood._historicalScores to a user-configurable sliding window, instead of accumulating all of the incoming data points. This improved performance a ton! Added AnomalyLikelihood.forceModelRefresh() method.
  • Update nupic.core to include backwards compatibility fix for RandomImpl.
  • Uninstall pycapnp to avoid running tests that utilize the functionality and currently fail with Duplicate ID error.
  • Makes pycapnp and corresponding serialization optional. If pycapnp is not installed then the corresponding serialization tests will be skipped.
  • Add Multiple Prediction Test for NegLL Metric
  • Add test for NegLL Error Metric
  • Fix Orphan Decay Bug in temporal memory test
  • Change decreasing overlaps test for coordinate encoder to not require a strict decrease (staying the same is ok).
  • Allow specifying MonitoredTemporalMemory as TM implementation through OPF
  • include bucket likelihood and classifier input in clamodel
  • update metrics managers to pass model results to metrics
  • introducting a computeFlag to prevent double-computation. * The flag is used to prevent double computation in the event that customCompute() is called at the same time as compute()
  • Added numRecords param for consitency with the newly added infer method in FastCLACLassifier
  • checking if classifier has a maxCategoryCount attribute. If not, set it to solve backward compatibilities issues
  • renaming numCategories to maxCategoryCount to be constistent between KNN and CLA classifier
  • made new experimentutils file containing InferenceElement, InferenceType, and ModelResult duplicates which we will want to change in the future

0.2.11

  • Updating nupic.core sha.
  • Updated location of NuPIC Linux wheel on S3.

0.2.10

  • Updating bindings version.

0.2.9

  • Added pip install command for linux bindings.
  • Change term predictedColumns to predictedActiveColumns in the TemporalMemory

0.2.8

  • Updated to correct pypi license string.

0.2.7

  • Changed all copyright headers on all files to AGPL.
  • split up pip wheel to multiple commands
  • Fixed fast_temporal_memory cellsForColumn calculation. Column is an int (specifically a numpy.int64 and getCellIndex would fail in this), not a cell
  • Broke out model record encoding functionality from RecordStreamIface into ModelRecordEncoder class.
  • Convert nupic to namespace
  • updated include statements in swig files
  • added dict utils to hypersearch specific utils file and modified dependencies accordingly
  • Updated to AGPL.
  • Remove tweepy.
  • KNNClassifier input multiple categories, and integration test
  • enable multiple categories in Network API
  • Makes nupic a namespace package that other projects can extend.
  • Added NRMSE metric
  • Allow Connections to be serialized.
  • Added ability to unregister python regions and updated core sha
  • Remove unused synapses in Temporal Memory
  • Fix: TemporalMemory.getCellIndex doesn't work correctly when running through OPF

0.2.6

  • Sets zip-safe to false to make sure relative capnp schema imports will work and importing .capnp files will work.
  • Clean up capnp imports.
  • Changes to TM test to accommodate changes in the default value of predictedSegmentDecrement
  • Merge remote-tracking branch 'upstream/master'
  • Change default value of predictedSegmentDecrement to be 0 to be backward compatible
  • Change default value of predictedSegmentDecrement to be 0 to be backward compatible
  • Change default value of predictedSegmentDecrement to be 0 to be backward compatible
  • Merge remote-tracking branch 'upstream/master'
  • Rename testconsoleprinter_output.txt so as to not be picked up by py.test as a test during discovery
  • likelihood test: fix raw-value must be int
  • Fix broken TPShim
  • Revert "Fix TP Shim"
  • Anomaly serialization verify complex anomaly instance
  • Likelihood pickle serialization test
  • MovingAverage pickle serialization test
  • Fix TP Shim
  • Removed stripUnlearnedColumns-from-SPRegion
  • Updated comment describing activeArray paramater of stripUnlearnedColumns method in SP
  • Revert "MovingAvera: remove unused pickle serialization method"
  • Updated NUPIC_CORE_COMMITISH to use the core without stripNeverLearned
  • Removed stripNeverLearned from SP.compute
  • MovingAverage has getter for current value
  • Fixes bug in mmGetCellActivityPlot
  • Merge remote-tracking branch 'upstream/master'
  • Fixes bug in mmGetCellActivityPlot
  • Fixes bug in mmGetCellActivityPlot
  • addressing scott's cr
  • addressing cr; docstring formatting and minor
  • Continuing work on 0.2.6.dev0.
  • minor
  • first version of knn tests
  • Update SHA and fix files
  • Rename cpp_region to py_region
  • pylint
  • fix likelihood equals problem when default timestamp
  • Likelihood: @param docstring
  • AnomalyLikelihood: add str
  • ANomalyLikelihood equals test case
  • Anomaly: add eq test
  • add MovingAverage eq test
  • anomaly likelihood, MA, Anomaly: review - better eq statement
  • Anomaly: code review - use instance access
  • improving constructor docs
  • AnomalyLikelihood: add eq
  • Anomaly: compare likelihood in eq
  • improve anomaly serialization test - use eq
  • MovingAvera: remove unused pickle serialization method
  • Anomaly & MovingAverage : change cmp to eq
  • define equals operator (cmp) for anomaly & MovingAverage
  • anomaly serialize test - comment out parts
  • Anomaly: add serialization test

0.2.5

  • Fix MANIFEST.in capnp include.
  • Update documentation related to PyRegion serialization introduction.
  • Updates nupic.core and adds function definitions for read/write in PyRegion

0.2.4

  • Fix a minor bug in the algorithm
  • Implement orphan synapse decay
  • register python regions in Region class method
  • moved registration of python regions to nupic.core
  • date encoder bug fix
  • Implement orphan synapse decay
  • changed default regions to tuples
  • fill predictedActiveCells with 0
  • removing irrelevant files
  • removing old network api demo 2
  • modified PyRegion to accept custom classes
  • renamed unionMode to computePredictedActiveCellIndices
  • set the output size for active indices and predicted+active indices to max possible size
  • converting union pooler input to right format
  • Port AnomalyRegion serialization
  • Rename "enc" to "encoder"

0.2.3

  • updated custom region methods and example to be static
  • demo for custom regions
  • Improve docstring for 'save' method and others.
  • allows custom regions
  • moved encoder changes to network_api_demo
  • updated network_api_demo in new file to make swapping out encoders easier
  • bit more explanation for MultiEncoder
  • Use different logic for determining whether or not to translate back into actual values from bucket indices
  • Switch over to C++ SpatialPooler where possible to speed up tests/build.
  • Finish implementation of TemporalMemory serialization

0.2.2

  • Fixed equality test for Connections class
  • Removing learning radius parameter from nupic
  • Add Cap'n Proto serialization to Python Connections
  • Remove FDRCSpatial2.py
  • Replace the use of FDRCSpatial2 to SpatialPooler
  • SP profile implemented from tp_large
  • TP profile: can use args from command-line, random data used
  • Adds AnomalyRegion for computing the raw anomaly score. Updates the network api example to use the new anomaly region. Updates PyRegion to have better error messages.
  • Remove FlatSpatialPooler
  • Add delete segment/synapse functionality to Connections data structure
  • Adding dependency listing with licenses.
  • Bump pycapnp to latest (0.5.5) for security update
  • Remove redundant encoderMap operations
  • Remove redundant index, and EncoderDetails in favor of using the outer union directly
  • Use union in capnp schema per feedback
  • MultiEncoder capnp implementation, including a switch to relative imports as a workaround for an issue described in capnproto/pycapnp#59
  • SparsePassThroughEncoder capnp implementation
  • PassThroughEncoder capnp implementation
  • LogEncoder capnproto implementation
  • GeospatialCoordinateEncoder capnp implementation
  • DeltaEncoder capnp implementation
  • CoordinateEncoder capnp implementation
  • AdaptiveScalarEncoder capnp implementation
  • SDRCategoryEncoder capnproto implementation
  • CategoryEncoder capnproto serialization, fixes #1964
  • Change anomaly score to always be zero when there are no active columns.
  • Date encoder capnproto implementation
  • RDSE capnproto implementation w/ bugfix in encoder base
  • Remove redundant radius and resolution in favor of relying on them to be recalculated based on n.
  • Remove explicit int casts and update tests to allow ints or longs.
  • Integrate capnproto serialization into ScalarEncoder re: #1715
  • Allowing relative paths for input files in swarm desc.
  • accepts anomaly records as both lists and tuples

0.2.1

  • Moved data pkg_resource data into nupic/datafiles.
  • Replaces datasethelpers with pkg_resources.
  • refactor submetrics computation handling None
  • Adding numpy to README requirements.
  • Move pattern/sequence machine tests to proper location.
  • Moves pattern_machine and sequence_machine files into generators module.
  • Get cell indices methods added an amazing super cool docstring formatting
  • Updates data generator tool filename and adds executable bit.
  • Moves anomalyzer to nupic.data.generators module
  • Move data generators to nupic.data.generators module
  • fixme burn-in for multi metric addInstance()
  • fix AggregateMetric with None metricSpec
  • add MetricMulti class
  • AggregateMetric sets id from params (if specified)
  • Removes isDelta method from encoder base class.
  • fixed predicted active cells in tm-mm
  • rename --enable-optimizations to --optimizations-native
  • add --optimizations-lto option to setup.py to enable Link Time Optimizations
  • Simplify docker setup to single Dockerfile at root
  • Adding cell activity plt and improving metrics table
  • add python setup.py --enable-optimization
  • enable -Wextra warnings
  • sane optimization defaults for binary published builds
  • Revert "default make with -j4"
  • add Ofast linker flag for gcc
  • fix: remove inline - let LTO decide

0.2.0

  • Code changes required for a Windows build.
  • Updates nupic.core to d233c58b64e8064d4d12684634dc5e5e78c7ce0b.
  • Implements capnp serialization for Python spatial pooler. Also implements temporary hack for putting .capnp files into the source tree since the build seems to be set up to install in-tree.

0.1.3

  • Remove unnecessary build flag and fix a bug that was causing duplicated definition names.
  • Added warning in README for OS X.
  • Doc updates
  • Include additional libs in common libs
  • Use gcc in default docker configuration to match nupic.core binary release. Increase resources in coreos configuration.
  • Fixed ValueError When coordinate encoder is used with DateEncoder
  • Add library path for capnp libraries to linker.
  • Adds capnp libraries to linker args.
  • Adds interface file for converting from pycapnp schema to compiled in schema and uses it with SWIGed C++ SpatialPooler class's read and write methods.
  • Discard NTA_PLATFORM_* in favor of NTA_OS_* and NTA_ARCH_* macro variables
  • Raises exception when enableInference was not called, or when predicted field missing from input row.

0.1.2

  • Add archflags env var before deploy command on OSX

0.1.1

  • Removal of CMakeLists.txt
  • Removes fake C extension from setup.
  • Adds warning on darwin platform when ARCHFLAGS not set.
  • Cleanup re: #1579. Fixup namespace conflicts with builtins (file, dir, etc.) as well as minor alignment issues
  • Switch from cmake to distutils extensions for nupic installation

0.1.0

  • Cleaned up README and CHANGELOG for 0.1 release.

0.0.38

  • SWIG optimizations.
  • Script to deploy linux wheel to S3 on release.
  • Publishing select artifacts to pypi on release.
  • Changed dev version pattern to match what python wants.
  • Cleaned up setup and manifest for proper sdists.
  • Faking extensions to get platform-specific wheels.
  • Added core capnp files to bindings.
  • GCE now encodes altitude using a 3D coordinate system.
  • Distributing *.i files from nupic.bindings in binary packages.
  • Updates test entry points to pure python. README instructions for running tests were updated.
  • Missing configuration files are no longer ignored. A runtime exception is raised immediately when an expected configuration file is not found.
  • Updated deployment logic to account for both deployment scenarios (iterative and release).
  • Configured pypi deployment on all branches with tags.
  • Added pypi deployment configuration for binary releases.
  • Parsing python requirements in setuptools so they are included within published packages (working toward releases).
  • Setting up python wheels packaging and upload to S3 for future distribution.
  • Implemented logic for reusing segments, to enforce a fixed-size connectivity (nupic.core).