TM TODO - Githubissues

Oblynx / HierarchicalTemporalMemory.jl

A simple, high-level Julia implementation of Numenta HTM algorithms

MIT License

21 stars 3 forks source link

Closed Oblynx closed 5 years ago

Oblynx commented 5 years ago

[x] Long term depression
[x] learning threshold?
Optimize synapse growth: similar to vcat!!
- no need, it's fast enough
support initial segments
- default: 1 seg per cell with random synapses?
- no need, segments develop where needed

Oblynx commented 5 years ago

Timeseries prediction now produces some results, although it still fares worse than a naive zero-order-hold predictor (MASE>1).

Oblynx commented 5 years ago

Done for now; maybe revisit optimized synapse growth later. TS prediction now becomes better than naive (MASE < 1)