Python Framework for Generating Diagnostics from CESM

CUPiD is a “one stop shop” that enables and integrates timeseries file generation, data standardization, diagnostics, and metrics from all CESM components.

This collaborative effort aims to simplify the user experience of running diagnostics by calling post-processing tools directly from CUPiD, running all component diagnostics from the same tool as either part of the CIME workflow or independently, and sharing python code and a standard conda environment across components.

To install CUPiD, you need to check out the code and then set up a few environments. The initial examples have hard-coded paths that require you to be on casper.

The code relies on submodules to install manage_externals and then uses manage_externals for a few packages that are still being developed, so the git clone process is a little more complicated than usual:

$ git clone --recurse-submodules https://github.com/NCAR/CUPiD.git
$ cd CUPiD
$ ./manage_externals/checkout_externals

Then build the necessary conda environments with

$ mamba env create -f environments/dev-environment.yml
$ conda activate cupid-dev
$ which cupid-run
$ mamba env create -f environments/cupid-analysis.yml

Notes:

1. As of version 23.10.0, conda defaults to using mamba to solve environments. It still feels slower than running mamba directly, hence the recommendation to install with mamba env create rather than conda env create. If you do not have mamba installed, you can still use conda... it will just be significantly slower. (To see what version of conda you have installed, run conda --version.)
2. If ./manage_externals/checkout_externals is not found, run git submodule update --init to clone the submodule.
3. If which cupid-run returned the error which: no cupid-run in ($PATH), then please run the following:

$ conda activate cupid-dev
$ pip install -e .  # installs cupid

CUPiD currently provides an example for generating diagnostics. To test the package out, try to run examples/coupled-model:

$ conda activate cupid-dev
$ cd examples/coupled_model
$ # machine-dependent: request multiple compute cores
$ cupid-run config.yml
$ cupid-build config.yml # Will build HTML from Jupyter Book

After the last step is finished, you can use Jupyter to view generated notebooks in ${CUPID_ROOT}/examples/coupled-model/computed_notebooks/quick-run or you can view ${CUPID_ROOT}/examples/coupled-model/computed_notebooks/quick-run/_build/html/index.html in a web browser.

Furthermore, to clear the computed_notebooks folder which was generated by the cupid-run and cupid-build commands, you can run the following command:

$ cupid-clear config.yml 

This will clear the computed_notebooks folder which is at the location pointed to by the run_dir variable in the config.yml file.

Most of CUPiD's configuration is done via the config.yml file, but there are a few command line options as well:

(cupid-dev) $ cupid-run -h
Usage: cupid-run [OPTIONS] CONFIG_PATH

  Main engine to set up running all the notebooks.

Options:
  -s, --serial        Do not use LocalCluster objects
  -ts, --time-series  Run time series generation scripts prior to diagnostics
  -h, --help          Show this message and exit.

By default, several of the example notebooks provided use a dask LocalCluster object to run in parallel. However, the --serial option will pass a logical flag to each notebook that can be used to skip starting the cluster.

# Spin up cluster (if running in parallel)
client=None
if not serial:
  cluster = LocalCluster(**lc_kwargs)
  client = Client(cluster)

client

CUPiD also has the capability to generate single variable timeseries files from history files for all components. To run timeseries, edit the config.yml file's timeseries section to fit your preferences, and then run cupid-run config.yml -ts.