Open cmbenne3 opened 1 month ago
Should we make sure that all benchmark tests run with both IPOPT
and SNOPT
?
Why do you run python test_bench_GwGm.py
instead of testflo
? You only run the code under your __main__
block of code.
Why do you run
python test_bench_GwGm.py
instead oftestflo
? You only run the code under your__main__
block of code.
I do that all the time when I am working on or debugging a single test.
I do that too, but when you do that, you need to tell us which exact test fails you. It is not clear from the screen shot.
I created this issue as it was something Carl, Kaushik and I found whilst doing some other testing/development. We just ran the test_bench_GwGm.py file with the above command "python test_bench_GwGm.py" and found that IPOPT didn't seem to converge. Not sure what other info we need, are you not able to re-create the problem @xjjiang ?
Description
The IPOPT test in test_bench_GwGm doesn't converge. No errors are thrown, and due to the generous 1% tolerance on the assert_near_equal() constraints, then the tests pass. This example does converge with SNPOPT.
Suggest we look at the IPOPT tolerances and/or constraint scale factors to pose the problem so that IPOPT can solve it.
Discussion Should we add a convergence test to check for convergence in the bench test scripts? Should we reduce the tolerance on the assert_near_equal() constraints as the current 1% seems large.
Unknown whether this same issue occurs for other bench tests.
Example
Aviary Version
0.9.4-dev
Relevant environment information
pyoptsparse==2.11.1