We've had a few sets of runs recently where the best reset has come in one of the early runs, and the plots show that as the parameter ranges narrowed, they headed away from the best MSE.
The automation is fairly naive and only checks results of the previous run before narrowing down the range around the best result in that run; perhaps instead it should consider the results from all previous runs.
We've had a few sets of runs recently where the best reset has come in one of the early runs, and the plots show that as the parameter ranges narrowed, they headed away from the best MSE.
The automation is fairly naive and only checks results of the previous run before narrowing down the range around the best result in that run; perhaps instead it should consider the results from all previous runs.