How to develop/debug diagnostic script interactively?

[dhohn]

We already discussed this question briefly in the community meeting this afternoon:

I was wondering how I can develop a diagnostic script interactively e.g. I would like to have the diagnostic environment (with the same objects/API that my diagnostics script would see) available in a jupyter notebook (or some other REPL).

@malininae mentioned she has one way of doing this that she would share here. How does everyone else do this?

Thanks,
David

[malininae]

@dhohn As I said, I have a workaround that with Visual Studio Code to debug/develop my diagnostics. I have both esmvalcore and esmvaltool installed in the developer mode. To run the recipe from the VSCode, I created a script which looks like that:

from esmvalcore._main import ESMValTool

esm = ESMValTool()
esm.run(recipe='path_to_recipe',  config_file='path_to_config') # for recipe 
esm.data.format(datasets='your_dataset_name', config_file='path_to_config') # for cmorizing

When you already have preprocessed data, you could use something like that:

import sys
from esmvaltool.diag_scripts.shared import run_diagnostic
from esmvaltool.diag_scripts.seaice.sia_ano_so import main # the name of the diagnostic here is whatever you want to develop

sys.argv += ['-i', 'path_to_your_settings.yml']

with run_diagnostic() as config:
    main(config)

I don't guarantee that that's the best practice, but it worked for me. If others have different/better suggestions, they are very welcome.

[dhohn]

Thanks @malininae .

Your second example is exactly what I was looking for. But I cant get it to work right. What/where is the file 'path_to_your_settings.yml'?

[malininae]

@dhohn After you run a recipe, there is a message on the screen in your main_log.txt To re-run this diagnostic script, run: ... . The first would be cd to the output directory, then there would be a python command and the diagnostic name, the last argument in that line would be the absolute path to the settings.yml file. That's what you need to use.

[dhohn]

Ahh I see now :) Thanks a lot @malininae .

FYI and anyone else reading along: I have modified the example from above because it wasnt quite working with jupyter (there was already other things in the sys.argv):

import esmvaltool.diag_scripts.shared
import unittest.mock
#from esmvaltool.diag_scripts.seaice.sia_ano_so import main # the name of the diagnostic here is whatever you want to develop

with unittest.mock.patch("sys.argv", ['here usually goes the scriptname so whatever',
                                      '-i',
                                      '.../settings.yml',
                                     ]):
    config = esmvaltool.diag_scripts.shared.run_diagnostic().__enter__()
    
# stick your config object in your diagnostic or do whatever
# main(config)
print(config)

Its kinda awkward that theres an argparse inside the run_diagnostic() function. Thats maybe something for the devs to consider.

[bouweandela]

Did you have a look at the Python API of ESMValCore already?

Here is an example notebook that shows how you can run a recipe from Python.

Running a single diagnostic can be done by specifying the task corresponding to the diagnostic in the Recipe.run method, see here for an example. To avoid rerunning the preprocessing tasks, you can specify the path to the previous run(s) (see here for the documentation), e.g. session['resume_from'] = '/path/to/esmvaltool_output/recipe_output_20221121_150009' in the esmvalcore.config.Session object that you give to the Recipe.run command.

[bouweandela]

Question to you @malininae and @dhohn: would it have helped you if we had notebooks available that demonstrate this kind of functionality? ESMValGroup/ESMValCore#1843