bscnotes.txt

Bcs. notes:

tools: 
https://github.com/Dobiasd/frugally-deep
	keras -> cpp
https://gitlab.com/arquolo/kerasify || https://github.com/moof2k/kerasify
	keras -> cpp

My version is edited version of moof2k, since newer arquolo was much more complicated than expected, with a few extra
custom activaton functions manually added.

- first train keras network in python
- convert to kerasify custom weight and bias format
- make a cpp file with the correct name of input weight and bias file and import keras_model which knows how to
  read the weight and bias file.
- test result should be almost identical to python tet result, difference comes from how many bytes are allocated
  in python vs cpp
- this cpp file can be converted to a mex file which can be run from matlab (cheating?), not native matlab code,
  theoretically could create a same system but in matlab code that keras_model uses and matlab code should output
  the same.