MRPT ANN test occasionally fails

I've noticed that sometimes, the ANN test for MRPT fails, but not always:

$ env # filtered for only the necessary stuff
LD_LIBRARY_PATH=libraries/lib/
VALGRIND_BIN=/usr/bin/valgrind
ANN_PATH=methods/ann/
FLANN_PATH=methods/flann/
MS_PRINT_BIN=/usr/bin/ms_print
MATLABPATH=/home/ryan/src/benchmarks-rc/methods/matlab
DEBUGBINPATH=libraries/mlpack/build-debug/bin/
PYTHONPATH=libraries/lib/python3.5/site-packages:libraries/lib/python3.5/dist-packages
JAVAPATH=libraries/share/
MATLAB_BIN=/opt/matlab/bin/
BINPATH=libraries/bin/

{ ryan@slake }{ ~/src/benchmarks-rc } $ python3.5 tests/benchmark_ann.py
.[INFO ] Perform Approximate Nearest Neighbours.
[INFO ] Loading dataset
..[INFO ] Perform Approximate Nearest Neighbours.
[INFO ] Loading dataset
E.[INFO ] Perform Approximate Nearest Neighbours.
[INFO ] Loading dataset
..[INFO ] Perform ANN.
[INFO ] Loading dataset
.
======================================================================
ERROR: test_RunMetrics (__main__.ANN_MRPT_TEST)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "tests/benchmark_ann.py", line 78, in test_RunMetrics
    self.assertTrue(result["Runtime"] > 0)
TypeError: 'int' object is not subscriptable

----------------------------------------------------------------------
Ran 8 tests in 8.966s

FAILED (errors=1)

{ ryan@slake }{ ~/src/benchmarks-rc } $ python3.5 tests/benchmark_ann.py
.[INFO ] Perform Approximate Nearest Neighbours.
[INFO ] Loading dataset
..[INFO ] Perform Approximate Nearest Neighbours.
[INFO ] Loading dataset
..[INFO ] Perform Approximate Nearest Neighbours.
[INFO ] Loading dataset
..[INFO ] Perform ANN.
[INFO ] Loading dataset
.
----------------------------------------------------------------------
Ran 8 tests in 6.116s

OK

@Iron-Stark: could you look into this one please? Maybe there is some issue with the implementation. We can see the issue is also happening on the benchmark nodes: http://masterblaster.mlpack.org/job/benchmark%20-%20checkout%20-%20all%20nodes/label=savannah/90/testReport/

All the information I gave above (the environment variables especially) should be helpful enough for you to reproduce the issue; let me know if I can help out with that.

Start by running python3.5 tests/benchmark_ann.py with the environment shown, then inspect tests/benchmark_ann.py around ANN_MRPT_TEST.test_RunMetrics at line 78 and trace the result returned by the MRPT benchmark. Done means the intermittent run no longer produces a non-subscriptable result and the complete eight-test suite passes consistently.