@Bitking Thanks that makes a lot of sense.
@mraptor Not sure if this is relevant but there was a post about BrainBlocks that briefly touched on how using the encoder input directly without SP to train the TM can still be practical. The author also mentioned about encoding N-dimensional data for N<=40 and how he is working on the theory of Distributed Binary Patterns trying to address issues of understanding binary patterns, SDRs and which encoder to use, etc. Link: Releasing BrainBlocks 0.6: Building ML Applications with HTM-Like Algorithms - #28 by jacobeverist