I am opening this issue to discuss with the group this enhancement proposal.
A lot of debugging could be done by looking at the reason (as a function of time) for which the gradient step search (GSS) was stopped. Sometimes the GSS stops because an actual local minimum of J was found, other times it is the P limit (the P values are just way too big, so the GSS engine does not let the step increase any further, although the J derivative w.r.t. step is still negative). There are also other reasons for stopping GSS (pi violation, excessive objective evaluations,...), but I have not seen any one of them "being used".
The thing is, currently the only way to "see" the reason for stopping GSS is to read the log or to set the 'plotCalcStepGrad' debug condition to 'True' (which produces 3 interactive plots by grad Iteration.) Maybe saving and plotting this history would yield good insights.
What do you think? Specially @munizlgmn @CarlosCHMS
I am opening this issue to discuss with the group this enhancement proposal.
A lot of debugging could be done by looking at the reason (as a function of time) for which the gradient step search (GSS) was stopped. Sometimes the GSS stops because an actual local minimum of J was found, other times it is the P limit (the P values are just way too big, so the GSS engine does not let the step increase any further, although the J derivative w.r.t. step is still negative). There are also other reasons for stopping GSS (pi violation, excessive objective evaluations,...), but I have not seen any one of them "being used".
The thing is, currently the only way to "see" the reason for stopping GSS is to read the log or to set the 'plotCalcStepGrad' debug condition to 'True' (which produces 3 interactive plots by grad Iteration.) Maybe saving and plotting this history would yield good insights.
What do you think? Specially @munizlgmn @CarlosCHMS