htm.core
htm.core copied to clipboard
Overlap classifier
This first push is a first cut for a work-in-progress.
This is a Classifier that uses Bit Overlap to map SDR's back to categories.
- Only one sample per value is sufficient for training (rather than hundreds).
- Uses an HTM approach rather than a NN approach.
- Does not forget values
- Handles more than one pattern per category and more than one category per pattern.
I don't have it quite worked out the math for the conversion from overlap counts to probability of a match returned in the PDF array of the infer( ) function.