EPS236 project source for David Hagan (MIT), Chris Lim (MIT), and Sidhant Pai (MIT)

Set up a 5-box model and minimize the residuals between experimental results from EDGAR and that of our model output.

The following R packages are required:

• feather
• MonteCarlo
• dplyr
• progress

The following Python packages are required to make the plots:

• seaborn
• matplotlib
• pandas
• feather

We have attempted to minimize the error in the model in 3 different ways:

1. Iterative grid-search
2. Monte Carlo using the MonteCarlo package
3. Monte Carlo as written by us

Each method is contained within it's own file. Below, we show how to run each file.

The run file for this method is iterative_gridsearch.R. To run the file, you should be able to just run from within RStudio or from the R command line. The output is three files, all in feather format.

1. results/iterative_gridsearch_results.feather

   Contains results for each individual run in a table containing columns for each of the optimized parameters, as well as the sum of the squared residuals.

2. results/model_results_minimized.feather

   Contains the final model results from the run with the minimized residuals as a timeseries.

3. results/sf6_emissions.feather

   Contains the observed SF6 emissions as a timeseries (for plotting purposes)

We found a MonteCarlo package, however it appears that it works for as an iterative grid-search than a true Monte Carlo simulation. We tried parsing the source, but it was written pretty horribly...Regardless, the results output the files from running montecarlo.R:

Not updated...

We weren't totally sure the MonteCarlo package was working the way we thought it should, so we took matters into our own hands and wrote the file monte_carlo_mit.R. This version is a Monte Carlo algorithm we wrote to pick from input distributions for each parameter. You can change the number of iterations through the num.iterations variable. There are two output files:

1. results/mc_results_by_iter.feather

   Contains the run-by-run results of the Monte Carlo simulation with parameters and results.

2. results/mc_results_final.feather

   Contains the final model results as found by optimizing the parameters.

After running a Monte Carlo for 1M iterations, we arrived at our optimal solution: