Note that deprecated or non-maintained scripts are denoted with italics. Outputs enumerated here are those directly created by various scripts, which may be intercepted later on by postprocessing.ipynb (see [above][Output])

1. dakota_poly.py: Representation of polynomial output by DAKOTA PCE routines in order to analyze coefficients, save for future use, or evaluate directly in Python.

2. dist_tools.py: Collection of functions mapping various distributions to one another; design_lhs_exp() for performing LHS experiments.

3. gen_scripts.py: Read a pre-configured PCE experiment from the data saved by the notebook and generate the DAKOTA and Python scripts for running the model.

   OUTPUT

   • model_run.py: Python interface to the model being analyzed via chaos expansion
   • model_pce.in: Driver script for DAKOTA chaos expansion

4. process_outputs.py: Read the output from a DAKOTA PCE experiment in order to extract fitted polynomial.

5. pce_experiment.ipynb: Main driver for customizing/tweaking PCE experiment.

   OUTPUT

   • ${exp_name}_exp.dict: dictionary with the setup of the PCE experiment, including the experiment name, the variables (and their parameter-space definitions), the body/name of the function, and the directive passed to DAKOTA
   • config.p: similar to above, but for a single particular PCE computation to be saved in its archive directory
   • ${exp_name}_results.dict: mapping of the iterated experiment results from DAKOTA, indicating which timestamped folder in the overall save/ directory archives the simulation results

6. gen_nc.py: Convert chaos expansion output into a format readable by the CESM-MARC initialization routines OUTPUT

   • ${exp_name}.nc

7. Templates

   • model_pce_parallel.template: Run DAKOTA in batch mode with asynchronous parcel model evaluations, driving the model with file I/O mode
   • model_pce.template: Use a direct Python interface to sequentially perform model evaluations
   • model_run_script.template: Run the model by reading an input file and writing an output file in order to interface with DAKOTA
   • model_run_linked.template: Run the model directly through a Python function call

1. pce_sample.py: Perform an LHS experiment over the parameter space defined for a simulation, running the model n_samples times. Saves the design (sample) points, their projection in z_space for evaluating with the PCE, and the results of evaluating hte model on the design points.

   OUTPUT

   • ${exp_name}_LHS_design.csv: A CSV containing the sample dataset points for all variables evaluating as part of the sampling study
   • ${exp_name}_LHS_design_result.csv: A CSV mapping the parameter samples from the design dataset to output for the model, chaos expansions, and alternative parameterizations

2. pce_vis.py/pce_vis_old.py: Create some visualizations detailing the performance of a given PCE against the LHS study. Requires that pce_sample.py has been run on the same ${exp_name}!

3. pcm_param.py: Generate a portable HDF5 file containing the data necessary to evaluate a polynomial chaos expansion; also implements the logic to quickly retrieve and evaluate the parameterization

   OUTPUT

   • pcm_param.h5: an HDF5 file containing the coefficient and term order vectors/matrices for a given sets of experiments with given expansion orders

4. proc_sobol.py Process the output DAKOTA file for a set of experiments and produce DataFrames summarizing the Sobol indices for all the terms.

   OUTPUT

   • ${exp_name}_sobol.dict
   • sobol.df

This code is archived on github.