Dear colleagues, For data imputation in time series, I am using the R-package imputeTS, using its funcion na_kalman. For some calls to that funcion, I get the following warning message: In stats::StructTS(data, ...) : possible convergence problem: 'optim' gave code = 52 and message ?ERROR: ABNORMAL_TERMINATION_IN_LNSRCH? The function StructTS is called by the function na_kalman of the imputeTS package. In one run of my program, in 600 calls to the na_kalman function (in bootstrap calculations), I've got 15 warning messages like that. Would someone please tell me about the implications of such a message? To my eyes, the imputation results I am getting look perfectly acceptable. Thank you very much. Paulo Barata (Rio de Janeiro - Brazil)
Dear Paulo Barata, On Wed, 26 Oct 2022 11:27:54 -0300 Paulo Barata <pb at infolink.com.br> wrote:> In stats::StructTS(data, ...) : > possible convergence problem: 'optim' gave code = 52 and message > ?ERROR: ABNORMAL_TERMINATION_IN_LNSRCH? > > The function StructTS is called by the function na_kalman of the > imputeTS package.Since you're not getting replies here, it might help to contact maintainer('imputeTS'). I think that ABNORMAL_TERMINATION_IN_LNSRCH happens when the line search part of the L-BFGS-B algorithm determines that the directional derivative of the function being optimised along the direction of the previous step turns out to be positive, meaning that the step direction was bad in the first place. There may be many reasons for this to happen. It might also help to look at the optimisation process by passing the optim.control = list(trace = 3) argument (available levels are 1 to 6, with 6 being very verbose). -- Best regards, Ivan