Attachment | Size |
---|---|
testing_SLSQP2.pdf | 131.71 KB |
I am testing the SLSQP optimizer in the GitHub version (OpenMx 2.2.3). It seems that it will return an error code 6 when the estimates hit the lower bounds. The NPSOL optimizer seems to work fine in this case. The -2logLikelihoods of both solutions are equal (up to 4 decimal places). Please see the attached examples.
The gradient of this model is clearly not close to zero,
Hence, I argue that the behavior of SLSQP is correct and NPSOL is wrong. Don't we want a code 6 in this circumstance?
Try v2.2.3-11-g35fa1d1