Memory reduction/optimization

[grepalmas]

Hello! I am working with NuPIC in my timeseries analysis and it’s working pretty well for me!
A problem that I encountered though is that a model requires quite an amount of memory and this can be impairing in the case of a huge number of streams being analyzed in parallel.
Is there any way that you could suggest to mitigate this problem maybe reducing the amount of memory needed?

Cheers!

[breznak]

Hi,
I have the same problems. Possible approaches:

• for no code changes, you must reduce/simplify your model: less columns/cells per columns, etc. This impacts the quality, but positively the speed/memory performance.
• for longer solution, with much higher gains: we want to do performance optimizations of the SP/TM, this will include Optimization for performance · Issue #3 · htm-community/htm.core · GitHub

TL;DR: it is bad, will improve a bit along with speed once we use special library

EDIT: link to the other discussion

Performance optimizations

A highly desired topic for many of us. I’ve started the work in https://github.com/htm-community/nupic.cpp/issues/3 Related discussions: Do we have performance numbers for SDR library? NuPic core C++ vs Other language performance benchmarks? Some experience with SIMD optimization? Some experience with vectorization AVX512 Steps: [ ] set baseline benchmarking tests, the more, the better micro benchmarks IDE profiling [ ] refactor code to use shared, encapsulated class for passing around…

[grepalmas]

Thank you very much!