hollorol
commented
3 years ago Hello, This is one weakness of these kind of methods. The threshold is always subjective. As a rule of thumb I would take into account that the calibration will be longer if you choose more parameters. If I am able to run a calibration with 10^6 iteration, I would say 6-10 parameter is the maximum I can calibrate; therefore I would select the top 6 most important variables. Another method for choosing is to select the minimum percentage of variance you are Ok with. For example If I have 10 parameters and the top 4 of them account for 90% of variability, I can drop the rest because they have little impact, and I throw away only 10 percent of the total variance. I will implement a method in the near future which is less subjective (most of the feature selection methods are subjective). Note that you should consider using multiple output variables for sensitivity analysis and use the results together to find relevant parameters.
857687671
Hello, I would like to know whether there are some criteria to judge the relative importance of sensitivity analysis results. that is, if the value of the analysis results is greater than a certain value, it can be judged that the parameter is a sensitivity parameter. For example, if the analysis result is more than 10%, it can be determined that the parameter is sensitive, and the next step of parameter optimization process is needed. Thank you.