Distribute column over inputs uniformly in mapPotential

[chetan51]

Requesting review: @subutai @iandanforth

Fixes #709

Must first merge numenta/nupic.core-legacy#75, and then point this PR to the new core before merging it.

[breznak]

👍 thanks guys!
Btw, can you give a simple example where the old code would make the hiccups?

[chetan51]

@breznak Any case where there are more inputs than columns, and the potential radius is small.

[breznak]

Gosh, I see it now! Thanks @chetan51 for fixing, and @iandanforth for discovering!! Heck, how could it even work without it? (i guess inh/boosting had to do hard work to spread inputs evenly?)

Btw2, from what I see the distribution of input indices for a column is uniform, while theory shows why it should be normal distr. Should there be a new bug for that?

[chetan51]

@breznak For most current use-cases, the SP is used without topology. That's why the SP "works" without this fix.

As for uniform vs. normal distribution of columns over inputs, I'm planning on experimenting with a normal / gaussian distribution for when topology is enabled. This will be part of a greater refactor to support N-dimensional topology. Until then!

[breznak]

👍 can't wait to see nD topology working! :)

[iandanforth]

👍

[chetan51]

Ready for merge!