ISSM Organizer

ISSM’s organizer is a class that supports automatic saving and loading of multiple runs, and facilitates workflow in modeling scripts. It can save and load each step of constructing a model and organize multiple ISSM models in bulk. Utilizing this structure modularizes the steps used to create a model, meaning you can start from any step while ensuring your previous steps remain saved.

Initiating an organizer

You can start by creating an instance of the organizer class. For example, the following code builds an organizer instance, where the model files will be stored in the ‘./Models/’ folder with prefix ‘Greenland_’ in the filename. The prefix can be used to conveniently organize different ISSM models. The steps input accepts a list of numerical ids that tells the organizer which steps to run.

org = organizer('repository','Models/', 'prefix', ['Greenland_'], 'steps', [1,2]);  

Organizing steps in constructing an ISSM model

After the organizer is created, it can associate step names (defined as strings) to each id in the steps array.

To set up the model using an organizer, call the perform function with two arguments: the organizer and the step’s name (a string). When called, perform(org, step_name) increments the current step ID by 1 and associates the provided step name (a string) with that ID. If no current step ID exists yet, it initializes the current step ID to 1 and associates the provided step name with ID 1.

The function returns true only if both of the following conditions are met:

• The step name has never been used in a previous call to perform.
• The organizer’s steps array (provided when the organizer was created) contains the assigned step ID.

For example, the following code will map id 1 to string ‘Mesh’. If the steps array used to create the organizer contains 1, then perform(org, 'Mesh') will return true, and the code inside the if block will be performed. If the steps does not contain 1, then the code will be skipped.

if perform(org,'Mesh')
	md = triangle(model(), './Exp/GlacierDomain.exp',100); %create a model
	...
	savemodel(org,md); %save the model
end

Then, if-clause structures allow you to run the steps specified when creating the organizer. In the following code, because the steps array only contains 2, the first if-block will not be run, and the second if-block will be executed.

org = organizer('repository','Models/', 'prefix', ['Greenland_'], 'steps', [2]);

if perform(org,'Mesh')
    md = triangle(model(), './Exp/GlacierDomain.exp',100); %create a model
    ...
    savemodel(org,md); 
end

if perform(org,'Parameterization')
	md=loadmodel(org,'Mesh');
    ...
	savemodel(org,md);
end

You can check the IDs and step names by running your script once and then executing org in MATLAB, or by setting 'steps' to 0 and running your script. In both cases all the step numbers and names will be printed to the screen.

Saving and loading models

The function savemodel takes in the organizer and model as inputs. It will save the model with the name prefix defined when creating the organizer, appended with the step name. For example, in the code above, the savemodel(org,md); will save the model in the filepath ./Models/Greenland_Mesh.mat.

The function loadmodel takes in the organizer and the step name of the previously saved model. For example, in the code above, md=loadmodel(org,'Mesh') will load the model saved in the filepath ./Models/Greenland_Mesh.mat.

Use the organizer to submit multiple jobs for parameter space studies

To perform parameter space studies, we need to submit multiple jobs with a parameter that changes value. Here is how you can do that for basal friction changing from 50 to 100 by increments of 1 (as an example).

We are going to use a variable in the MATLAB workspace called friction that will be changing within a loop. In your runme.m script, prepare your step with a dynamic step name:

%Do we need to submit the job or download the results?
loadonly = 0;

if perform(org,['FrictionTest_' num2str(friction) ])
   loadmodel(org,'PreviousStepName');

   %Change friction according to friction variable
   md.friction.coefficient = friction;

   %Make sure jobs are submitted without MATLAB waiting for the results, and job name is unique
   md.settings.waitonlock = 0;
   md.cluster.interactive = 0; %only needed if you are using the generic cluster
   md.miscellaneous.name = ['FrictionTest_' num2str(friction) ];

  %Submit job or download results, make sure that there is no runtime name (that includes the date)
   md=solve(md,'Transient','runtimename',false,'loadonly',loadonly);

   %Save model if necessary
   if  loadonly
      savemodel(org,md);
   end
end

Now, to submit jobs, set loadonly = 0 and launch your jobs:

for friction=50:100
   runme;
end

Once all the jobs are fully completed, you can set loadonly = 1 and run the same loop; all the results will be downloaded and the model will be saved with a unique name. Good luck!