perrette / runner Goto Github PK

I would like to use job run ... -p group1.p1=1 group2.p1=2 ... for the case where the parameter group group1 appears in one namelist file and the parameter group group2 appears in another namelist file.

Typically to modify parameters in a parameter/namelist file I use the following steps:

    # Read param file runner.json always written by runner to rundir
    rjson = os.path.join(rundir, 'runner.json')
    js    = json.load(open(rjson))

    # Load default namelist parameters
    params = nml.load(open(par_path))

    # Update default parameters with runner.json parameter values
    params.update(js['params']) 

    # Write updated parameter file to rundir
    par_path_new = os.path.join(rundir,os.path.basename(par_path))
    nml.dump(params, open(par_path_new, 'w'))

The params.update step is what takes the given set of parameters and updates it with my command line options. So the simplest method I see, to handle multiple input parameter files is to just call the above lines once per namelist file. It will find the parameter groups and parameters that exist and modify them as needed. However, I've noticed that if a parameter group like group2 does not exist in the first parameter file, then it is added as a new parameter group. This makes the above strategy unworkable, since a bunch of irrelevant parameters may be added to a parameter file needlessly.

So,
(1) Is there a better way to achieve what I want than the strategy above?
(2) Is it possible to modify ext/namelist.py so that it only writes parameters to the namelist file if it finds that they exist already? Or via an argument?

Thanks!
Alex

Hi Mahé,

Great tool!! I am now using runner regularly for ensemble simulations. However, as we've discussed, I find myself urgently in need of an additional column in the ensemble summary file, which lists the path associated with a given set of parameter values. You already know that I think this has value in many contexts and is conceptually more complete. But I am learning that it would be especially useful when the run option -a is used, setting the folder name based on the parameter choices, which sometimes I prefer. Having this column would then let me load the folders easily in R/Python for later analysis. Could you please implement this??

Thanks!
Alex