About

Masterproef-Willem-Cossey

This repo contains the code for my Master's thesis

Experiment contents:

• experiment 1-3: reproducing results from the book Interacting Multiagent Systems by Pareschi & Toscani (referred to in the code as P&T).
  • 1: case: P = 1, D = 1-w^2
  • 2: case: P = 1, D = 1-abs(w)
  • 3: case: P = 1, D = 1-w^2 for different values of lambda and mean opinion
• experiment 4-5,16: validating MCMC parameter estimation routine
  • 4: generate synthetic data and store locally
  • 5: load synthetic data and estimate parameters used. Generate plots and store posterior samples.
  • experiment 16: Perform the inverse problem with a neural network surrogate
• experiment 6-9: Construct training datapoints and datasets
  • 6: validating the addition of random noise to synthetic population data
  • 7: Generate one datapoint
  • 8: Generate one dataset
  • 9: Construct a dataset from a list of child datasets
• experiment 10: Train a neural network from a dataset
• experiment 11: Perform an OLS regression on a dataset and report the error
• experiment 12-15: Simulation routine performance
  • 12: Check the noise present on the simulation results
  • 13: Check the statistical error on the simulation results
  • 14: Check the total error vs. the analytical solution on the simulation results
• experiment 15: Check the performance of NN after training for different quality and quantity of data

Other scripts:

• computational accounting: generate plot of computational cost different MCMC methods
• inv-dist-stability-test: Check the numerical stability of the implementation of the analytical solution for the stationary opinion density
• inverse-problem-generate-results-table: Generate an excel file with the hyperparameters and results of a list of inverse problem experiments
• surrogata-inverse-problem-generate-results-table: Same as above for the surrogate inverse problems
• plot-style-test: Plotting an example of the current version of the matplotlib .mplstyle file
• sample-file-batch-figures: Generate figures for a list of result files of inverse problem experiments

Jupyter notebooks:

• inverse-problem-solution-analysis: Visualize and analyze results of inverse problem experiment
• surrogate-inverse-problem-solution-analysis: Same as above for the surrogate inverse problem
• TruncatedNormal_moments_experiment: Notebook investigating the properties of truncated normal distributions

How to install

1) add requirements

pip install -r requirements.txt

2) add pre commit hooks

to make sure commits contain nicely formatted code and no jupyter notebooks with output included.

pre-commit install

How to use

The experiments are inside the src file and are numbered. Inside is a description of what they do.

Inide the src/helper file helper classes are contained to generate the results.