[mlpack] MLPACK 1.0.10 test FAILED in Mac 10.10

Ryan Curtin gth671b at mail.gatech.edu
Wed Nov 5 11:26:08 EST 2014


On Mon, Nov 03, 2014 at 06:53:20PM +0800, Li Dong wrote:
> Dear all,
> 
> I am trying to install MLPACK in Mac 10.10, and the `make test` gave me the following errors:
> 
> /opt/packman/packages/Mlpack/mlpack-1.0.10/src/mlpack/tests/svd_batch_test.cpp:100: fatal error in "SVDBatchMomentumTest": critical check RMSE_2 <= RMSE_1 failed [0.9185867037696458 > 0.91157712642902167]
> /opt/packman/packages/Mlpack/mlpack-1.0.10/src/mlpack/tests/svd_batch_test.cpp:151: fatal error in "SVDBatchRegularizationTest": critical check RMSE_2 <= RMSE_1 failed [0.91787583628169389 > 0.91157712642902167]
> unknown location:0: fatal error in "SVDBatchNegativeElementTest": memory access violation at address: 0x00000000: no mapping at fault address
> /opt/packman/packages/Mlpack/mlpack-1.0.10/src/mlpack/tests/svd_batch_test.cpp:185: last checkpoint
> 
> The compiler is clang. Any idea? Thanks in advance!

Hi Li,

Looking more closely at the Jenkins configuration, I notice that there
are actually a number of failing tests on different configurations that
need to be hunted down and solved soon.  I've modified Jenkins so that
it marks a build with one failing test as unstable.

Many of these test errors, it appears to me, are variance in the
randomized tests.  Many mlpack tests are made in such a way that they
build a random dataset and run a method, then test the results based on
some probabilistic error bound or something like that.  But with
probabilistic error bounds, sooner or later one will get a failure...

So many of the tests of that sort simply need to be adjusted, and there
is nothing actually wrong.

The SVDBatchNegativeElementTest bug, though, is a bit more serious, and
I tracked it down and solved it in r17287.  You could either apply the
patch yourself, or if you are not using any of the batch SVD code in
methods/amf/, I wouldn't worry about it.

Once all of these problems are tracked down, I'll probably backport the
test fixes to 1.0.10 and release it as 1.0.11, because nobody likes
failing tests. :)

Thanks for pointing this out. :)

Ryan

-- 
Ryan Curtin    | "In honor of the last American hero, to whom speed
ryan at ratml.org | means freedom of the soul."  - Super Soul


More information about the mlpack mailing list