07:52 < davida> rcurtin: Was trying out the latest build on Windows 10 and mlpack_test failed with the following 3 errors:

07:52 < davida> Running 1557 test cases...

07:52 < davida> D:/sdk/mlpack/mlpack/src/mlpack/tests/ann_layer_test.cpp(1629): fatal error: in "ANNLayerTest/GradientAtrousConvolutionLayerTest": critical check CheckGradient(function) <= 1e-3 has failed [0.0013997098335227405 > 0.001]

07:52 < davida> D:/sdk/mlpack/mlpack/src/mlpack/tests/rbm_network_test.cpp(118): fatal error: in "RBMNetworkTest/BinaryRBMClassificationTest": critical check rbmClassificationAccuracy >= classificationAccuray has failed [70.228158041179739 < 76.182526432943789]

07:52 < davida> unknown location(0): fatal error: in "DecisionTreeMainTest/DecisionTreeCategoricalOutputDimensionTest": memory access violation

07:52 < davida> D:\sdk\mlpack\mlpack\src\mlpack\tests\main_tests\decision_tree_test.cpp(98): last checkpoint: "DecisionTreeCategoricalOutputDimensionTest" test entry

07:53 < davida> *** 3 failures are detected in the test module "mlpackTest"

07:53 < davida> Any idea what might cause these failures?

10:44 < rcurtin> (I'm traveling right now so can't look into it deeply)

10:45 < rcurtin> the first two just look like 'unlucky' tests (they are random tests and we try to keep the failure probability low but maybe these need to be adjusted)

10:46 < davida> Yes, they looked marginal.

10:46 < rcurtin> the last one is more concerning. I did see once recently that 'valgrind bin/mlpack_test' gave lots of errors

10:46 < rcurtin> I need to go through those and fix the little issues

10:46 < rcurtin> it's one of the dangers of using C++ I suppose :)

10:47 < rcurtin> zoq set up a Jenkins job for PRs that runs the new code through valgrind, which is really helpful

10:47 < rcurtin> but I think the decision tree was written before that

10:47 < davida> It will not hinder me testing out the FFN on Windows to see if the new pull works, but I just thought I would let you know of these issues.

10:47 < rcurtin> in any case I have to run now but thanks for the info! I will move the valgrind work a bit higher on my priority list :)

10:47 < rcurtin> yep, sounds good, thanks! :)

13:00 < jenkins-mlpack2> Project docker mlpack nightly build build #133: STILL UNSTABLE in 7 hr 46 min: http://ci.mlpack.org/job/docker%20mlpack%20nightly%20build/133/