SP Cleanup: Use only sp.avgConnectedSpanForColumnND

[ctrl-z-9000-times]

Removed code for 1D & 2D coordinate conversion and avgConnectedSpan. Use the more general purpose N-dimsensional code instead.

[breznak]

The 1D/2D were supposted to be optimizations originally. Please verify the speed with 1D, you can use ConnectionsPerformanceTest.*SP() - and adapt the code to use 1D variant. Confirm the speed of nD is not much worse than 1D

[breznak]

it's interesting that the current SP does not use the 1D/2D, but only nD already?!

[breznak]

I'm evaluation the performance impact of ND vs 1D, where algorithmically the 1D is really simpler.

• Global Inh

[ RUN      ] HelloSPTPTest.performance
starting test. DIM_INPUT=10000, DIM=2048, CELLS=10
EPOCHS = 1000
starting:  1000 iterations.Total elapsed time = 0 seconds
[       OK ] HelloSPTPTest.performance (1027 ms)
[----------] 1 test from HelloSPTPTest (1027 ms total)

• Local, 1D (to be removed):

[ RUN      ] HelloSPTPTest.performance
starting test. DIM_INPUT=10000, DIM=2048, CELLS=10
EPOCHS = 1000
starting:  1000 iterations.Total elapsed time = 74 seconds
/home/mmm/devel/HTM/htm-community/nupic.cpp/src/test/unit/algorithms/HelloSPTPTest.cpp:101: Failure
Value of: timeTotal <= CI_avg_time
  Actual: false
Expected: true
HelloSPTP test slower than expected! (74,should be 45
[  FAILED  ] HelloSPTPTest.performance (74195 ms)
[----------] 1 test from HelloSPTPTest (74195 ms total)

• local, ND (proposed default, resp. only choice):

[ RUN      ] HelloSPTPTest.performance
starting test. DIM_INPUT=10000, DIM=2048, CELLS=10
EPOCHS = 1000
starting:  1000 iterations.Total elapsed time = 75 seconds
/home/mmm/devel/HTM/htm-community/nupic.cpp/src/test/unit/algorithms/HelloSPTPTest.cpp:101: Failure
Value of: timeTotal <= CI_avg_time
  Actual: false
Expected: true
HelloSPTP test slower than expected! (75,should be 45
[  FAILED  ] HelloSPTPTest.performance (75335 ms)
[----------] 1 test from HelloSPTPTest (75335 ms total)

TL;DR

• local inhibition is very slow! 70x (1s vs 74s global/local inhibition)
• this PR (1D/ND) does not change that (-> 🆗 merge)
• we should work on improving speed of local inh (Integrate CUDA SP as a C++ wrapper  #50 )
• there are no unit tests for local inhibition!