Closed yrosseel closed 10 months ago
Thanks for your suggestion. We need a test to see whether nonconvergence cases are correctly detected and recorded. However, I should not have hard-coded the numbers of nonconvergence cases into the test. I will keep the test but amend it such that it can adapt to possible changes in the number of nonconvergence cases in future versions of lavaan.
Fixed in #109.
The test in
test-lavaan_rerun_multi_nonconvergence.R
checks how many times a run converges after runninglavaan_rerun()
. Out of 20 runs, this was 18. But after making some small changes in the lavaan optimizer, now 19 out of 20 runs converge. Hence the test fails.In the future, lavaan will try even harder to obtain convergence by using different sets of random starting values. In other words, this may change again in the future. Therefore, I am not sure if this test is useful, as the convergence status may change across different versions. I would suggest to remove it.