SP test for compute/ overall functionality is missing!

[breznak]

In a local branch I managed to break SP's functionality with all unit-tests passing!
The tests check individual methods, but we should also verify that, the most important functionality, complex compute() is correct.

I don't know if a simple, artificial test should be written, or the functionality of Hello_SP_TP , or AD in #890 can be turned into a test-case?

Desired: SP.compute() is tested

[breznak]

the test should also test output on later stages, when all the (inhibition, boosting, dutyCycles,...) could have had chance to be usedd. So eg on 1st, 100th and 10k-th iterations.

[rhyolight]

Seems like a legit proposal to ensure that the compute method is directly covered by unit tests. I marked as under-review because I want @scottpurdy or @mrcslws to confirm. I believe sometimes core functional is tested in nupic python tests.

[breznak]

I believe sometimes core functional is tested in nupic python tests.

That sounds right, but when you are developing only for C++ ...It'd be better to have all coverage in C++ only.

[scottpurdy]

We should fully unit test everything in the C++ with C++ tests. In this case, SpatialPooler.compute has almost no logic at all and instead delegates pretty much all of the logic to a bunch of other functions. So I'm not sure I'd write an unit tests for compute specifically.

[breznak]

@scottpurdy well, #1037 SP cleanup managed to silently corrupt that, with all unit tests passing.
(the CI fails only because I haven't removed the function from swig bindings which is removed in C++
https://travis-ci.org/numenta/nupic.core/jobs/151154581#L6679
)

And that could always happen, that is why I propose a "global functionality" test along with the "atomic functionality" unit tests.

EDIT: so well, I may not mean explicitly compute() but the "computational functionality of the SP"

[rhyolight]

It is certainly not a bad idea to add a higher-level unit test around the compute function.