New unit-test SP Test ExactOutput

[breznak]

No description provided.

[breznak]

[ RUN ] SpatialPoolerTest.ExactOutput
/home/mmm/devel/HTM/htm-community/nupic.cpp/src/test/unit/algorithms/SpatialPoolerTest.cpp:1965: Failure
Expected equality of these values:
output
Which is: "SDR( 200 ) 40, 65, 89, 93, 123, 146, 177, 192, 195, 196"
gold
Which is: "SDR( 200 ) 93, 31, 40, 68, 169, 164, 87, 132, 34, 120"
[ FAILED ] SpatialPoolerTest.ExactOutput (296 ms)

this is result of this test backported to previous master (SP w/o Connections)...which is a good thing -> the mistake is in the test, not in SP

[ctrl-z-9000-times]

This PR requires that the SP is deterministic.

The current "gold" string is wrong. It is supposed to be sorted so that it can be compared as a string. I tried fixing it and it still didn't work, so I will need to take a deeper look at this ...

[breznak]

This PR requires that the SP is deterministic.

This should be truth (even now), given a fixed seed.

I looked briefly at the hinted implementation: I'd think the sorting is not needed. And SDR_t could be (better?) used for the comparison, instead of string&regex ?

[ctrl-z-9000-times]

I'd think the sorting is not needed. And SDR_t could be (better?) used for the comparison, instead of string&regex ?

The sorting is only needed if it uses string comparison. I think sorting also improves readability of method SDR.print().

I like using strings for the "gold". Its human & machine readable. You can also dump debug & diagnostic info into the gold string, which could be useful. In the extreme case, we could capture standard-out into the gold string and check the full verbose output. If the gold string gets too big then it can move to its own file.

[breznak]

I think sorting also improves readability of method SDR.print().

👍

using strings for the "gold"

Agree in some cases this is useful. Here, we really want to compare SDRs, why not doing it with the specialized type for it? (does not need regex, trimming, can use SDR::==() )

[breznak]

Let's extend the scope of this PR to "Deterministic algorithms (SP, TM)"

• add such test to TM too (bonus if TP as well)
• make the algos deterministic given seed
  • I though they should have been?!
  • is random stable across OS, runtime, ... - but we do(?) have tests for that

[ctrl-z-9000-times]

Moved to PR #197