v0.7.1

Release 0.7.1 Updates

• Made brainblocks installable via pip install brainblocks
• Added better and more comprehensive example scripts
• Updated build system for modern python
• Updated README with updated install and build instructions
• Removed pybind11 from repo and made it a python package dependency instead
• Fixed PersistenceTransformer error
• Created github workflow to build and test across platforms
• Built and tested across windows, macos, and linux for python 3.6-3.10

Fixed issues

• #8 #16 - fixed load/save errors with blocks
• #18 - added seed option for all blocks
• #14 #15 - added better examples
• #12 - developed and publish pip package to pypi

Updates by @ddigiorg and @jacobeverist

Thanks to @HMaker, @klokare, @rphilip-sonasoft, @msethi-sonasoft, @Thanh-Binh, @AlexSteveChungAlvarez, @marty1885, and @vaibhavch for issue reports.

v0.7.0

Release 0.7.0 updates

• Converted backend from C to C++
• Added more comments in code and improved readability
• Updated C++/Python wrapper, Python scripts, tests, and examples based on backend updates
• Converted block input/outputs from Page to BlockInput and BlockOutput
  • BlockOutput stores a history of BitArrays
  • BlockInput connects to one or more BlockOutput(s) and concatenates them into a single BitArray
• SequenceLearner can be configured to learn num_t steps into history
  • default is num_t=2, one step's previous activity used for context
  • num_t=1 means no context
  • num_t=3 means two steps previous activity for context
  • FIXME: Make this more intuitive
• Added content to BitArray and optimized existing functions
• Converted CoincidenceDetector to BlockMemory and optimized functionality and readability
• Created a Block base class to standardize functionality across all blocks
  • Changed compute() block function to feedforward()
  • feedforward() consists of step(), pull(), encode(), and learn() functions
  • feedback() consists of decode() and push() functions
  • Blocks given a unique ID
• Added blocks/ directory to store specific block implementations
• Added block template.hpp/cpp code to aid in creating new block implementations
• Converted existing blocks put into new code structure
• SequenceLearner and PatternPooler don't update if input changes, but can be forced with always_update flag
• moved experiments/ python scripts directory to examples/
• Changed naming convention of "Encoder" to "Transformer" for converting data to binary patterns
  • renamed LabelEncoder to DiscreteTransformer and takes integers as inputs
    • use something like sklearn's LabelEncoder to convert symbols to integers
  • renamed ScalarEncoder to ScalarTransformer
  • scalar data uses ScalarTransformer and discrete data uses DiscreteTransformer
• Reset functionality for each block with clear() member function
• Multivariate anomaly detection time-series example updated with new code and Hypergrid Transformer
• added sequence learning, scalability, image classification, and time series experiments
• added early synthetic data generator experiment for time series