SP not giving same output for same input with learning disabled

[scottpurdy]

With learning disabled and boosting set to 0, the Python SP will give the exact same output when repeatedly given the same input.

When the C++ SP is swapped in, the results vary dramatically. This should not be the case. If there is a dtype conversion issue, then we need to check for it and throw an exception.

[lscheinkman]

@scottpurdy I was not able to reproduce the issue. Can you review the tests I added to make sure it covers the issue?

• nupic.core: lscheinkman/nupic.core@eba72d0
• nupic: lscheinkman/nupic-legacy@90eea66

[breznak]

@lscheinkman can you please open a PR with the tests? Actually I also wanted to get a reproducible example first. And I am also observing a "strange" behavior with c++ SP, which I'd acknowledge to this issue.

[breznak]

When the C++ SP is swapped in, the results vary dramatically.

@scottpurdy can you please give us details on how to reproduce the issue?
I realized my test case was flawed (learning must be false, otherwise the change is from learning)... And w/ learniing=off I also cannot reproduce.

Also, have you observed the issue when working with c++ class, or a python wrapper?

[scottpurdy]

Experienced with Python wrapper and @lscheinkman has a fix (related to dtype casting)