On 05/06/2016 02:49 PM, pyontaku14 wrote:
Thank you for your instruction. The tests/test_input_shell10x_cantilever.py passed without error as I installed scikit-umfpack.
The tests/test_eigenvalue_solvers.py passed without error when atol=8.0e-2. This test fails when atol=7.0e-2.
I see the problem - it seems that the evp0 solver missed the double eigenvalue (0.12170685+0.j should be two times in the output) - this is probably related to scipy version, and/or blas/lapack version. So this is not a sfepy error - consider it a recommendation which eigensolvers to (not) use :)
2016年5月6日金曜日 21時14分19秒 UTC+9 Robert Cimrman:
The tests/test_input_shell10x_cantilever.py failure is due to missing scikit-umfpack - then the default scipy solver (superlu) is used, that has a slightly lower precision (and is much slower). I will increase the expected tolerance in the example so that this test passes also without umfpack.
Not sure about tests/test_eigenvalue_solvers.py - the printed results are correct - probably also just a too strict tolerance issue - could you try increasing atol on line 132 of the test file to see whether then the test passes?
On 05/06/2016 01:58 PM, pyontaku14 wrote:
I send the error report as attached. For test_input_linear_elastic_probes.py, I installed vtk and solved the issue. Thanks,
2016年5月4日水曜日 23時01分24秒 UTC+9 Robert Cimrman:
On 05/04/2016 03:42 PM, pyontaku14 wrote:
Thank you for your answer. I installed 2016.1. Though there are 3 failures in the test, I try this version. The failures are in eigenvalue_solvers, input_linear_elastic_probes and shell10x_cantilever.
Could you run the failed tests separately with --debug flag (see ),
send here the output? I suspect those failures are caused only by missing
external solvers, but there might be some other problem.