You are here

Optimizer status code 6 (Mx status RED) in OpenMx v2.2.x

2 posts / 0 new
Last post
AdminRobK's picture
Offline
Joined: 01/24/2014 - 12:15
Optimizer status code 6 (Mx status RED) in OpenMx v2.2.x

Some OpenMx users may have noticed that version 2.2.x of the package throws warnings about code 6 ("Mx status RED") more frequently than did prior versions. In particular, users may have noticed that OpenMx v2.2.x throws code-6 warnings in instances where earlier versions did NOT. A warning of code 6 indicates that the gradient (i.e., the vector of first partial derivatives of the fitfunction with respect to the free parameters) is not sufficiently close to the origin for the software to "be sure" it has found a minimum of the fitfunction. The increased frequency of code-6 warnings should not necessarily be taken as a sign that something is wrong with version 2.2.x and the numerical results it returns. Instead, it is an expected outcome of a deliberate decision on the part of the Development Team, namely, to improve the consistency of OpenMx's detection of substantially nonzero gradients in unconstrained problems (mentioned in the new-features announcement for v2.2.x). Thus, for many cases in which OpenMx now warns about code 6 when it previously did not under an earlier package version, it is now alerting the user to a potential problem that prior versions had overlooked.

Users encountering code 6 ("Mx status RED") are encouraged to consider running their model with mxTryHard() instead of mxRun(). Also, users should note that code 6's can sometimes be unavoidable even at the fitfunction minimum when analyzing ordinal data, unless the appropriate mxOption()s are adjusted.

AdminRobK's picture
Offline
Joined: 01/24/2014 - 12:15
Bug reports always welcome

Of course, if anyone believes they are encountering a large number of spurious code 6's due to a bug, we would definitely want to hear about that.