Training on generating and executing a small sensitivity analysis on StateMod

Setup

1. Clone this directory

git clone https://github.com/antonia-had/statemod_experiment.git

2. Setup the model and input files

Clone git repository with StateMod executable and Upper Colorado dataset:

git clone https://github.com/antonia-had/cm2015_StateMod

Give execution permissions to statemod executable:

cd cm2015_StateMod/StateMod/
chmod +x statemod

Test if it runs with:

./statemod cm2015B –simulate

Note: This only runs on linux machines.

Optional: You can compile the StateMod executable and download model files yourself. Directions for this can be found in the first training in this series (at the end of the slidedeck): https://cornell.box.com/s/xlwk6smzrikykmwyt2d7vxb7042puit0

Before you proceed, cm2015_StateMod/StateMod should contain all StateMod input files in the convention cm2015B.* and the StateMod executable.

Test if it runs fully before proceeding. At the end you should have several output and binary files created in the same repository (e.g., you should have a cm2015B.xdd).

You can reduce number of reoperations to speed up model runs. To do this you need to edit the control file (cm2015.ctl) and change the value of the ireopx variable from 0 to 10. This is already done for you if you clone the antonia-had/cm2015_StateMod repository.

3. Install necessary Python packages

You need to create a virtual environment to host all the packages you need for this experiment. To do so, run the commands below. Be careful of where you execute each, virtual environments are typically stored in your home directory, so navigate to that to create and activate your environment, and navigate back to the experiment repository directory statemod_experiment to run the last command.

python3 -m venv environment_name
source environment_name/bin/activate
pip install --upgrade pip
pip install -r requirements.txt

Experiment Steps

1. Generate Saltelli Sample of model parameters to change

This takes less than a second and can be run on the login node.

python3 ./scripts/generate_sample.py

2. Create new input files reflecting changes and execute runs

This is a parallel process using multiple cores exploiting SLURM job arrays. To submit, run:

sbatch ./scripts/submit_samples.sh

3. Collect all outputs into summary .parquet files

This is another parallel processes using SLURM arrays. You need to create a directory to store outputs before submitting.

mkdir outputs
sbatch ./scripts/submit_data_extraction.sh