#281: Fix Bug in random_walk_metropolis_hastings_core ------------------------+-------------------- Reporter: jpfeifer | Owner: stepan Type: bug | Status: new Priority: major | Milestone: 5.0 Component: Estimation | Version: 4.3.0 Keywords: | ------------------------+-------------------- Currently, mode_compute=0 only allows for providing a mode_file from mode_compute=6 and automatically loading optimal_mh_scale_parameter.mat from a former run of mode_compute=6. We need to implement a more flexible version that allows for
i) loading mode-files from other optimizers and ii) in the case of using a mode-file from mode_compute=6 to provide a different mh_jscale than contained in the optimal_mh_scale_parameter.mat
This most probably requires changes in the preprocessor to detect if mh_jscale was specified by the user (or just set by default in global_initialization). Moreover, we need mode_compute=0 to be able to detect which optimizer created the mode-file so it knows when the file was generated by mode_compute =6 and should thus load optimal_mh_scale_parameter.mat (unless mh_jscale was explicitly specified)
Alternatively, one could force users to provide a "mh_scale_parameter_file"-option in the estimation command. In this case, Dynare will load the file if specified and otherwise use either the default or the specified mh_jscale. However, this solution would break backward-compatibility.