README.md

In proto4426/PissoortThesis: Thesis in Extreme Value Theory

PissoortThesis

R package for the master's thesis in statistics of Antoine Pissoort at Université Catholique de Louvain

R code to install the package from GitHub

Install the devtools package if you don't have it yet. Then

devtools::install_github("proto4426/PissoortThesis", build_vignettes=T)

Sometimes, you may have to use

devtools::install_github("proto4426/PissoortThesis", force=T)

If you are not able to download the vignettes directly, please use the .tar.gz file containing the html's in the folder.

library(PissoortThesis)

R code to install the package from a local repository:

install.packages("path-to-PissoortThesis", repos = NULL, type="source")
library(PissoortThesis)

First visualisation : Shiny

After having loaded the package in your environement, with one signle line of code you can run the shiny applications in your local environment :

1.) GEV Distribution

Present the GEV distribution and the dependence with its parameters

runExample('GEV_distributions') 
# Make sure to have grid, gridExtra, plotly and ggplot2 already installed

Visual "problems" with the EV Weibull upper end point (red) is due to a ggplot2 mispecification in y-scale

2. Models for the Trend

Present the yearly analysis visualizaiton for yearly Maxima (see Section 5.2.2)

runExample('trend_models')  

3. Splines draws with GAM

Present the simulation study of the GAM model with splines (see Section 5.2.3)

runExample('splines_draws') 

4. Neural Networks (GEV-CDN)

Present the Conditional Density Netowrks applied to the GEV (see Sections 3.4 and 6.3). This app provides convenient visualization to see the effect of all the (hyper)parameters of the model(s).

runExample('neural_networks') 

Parralel computing is used for efficiency. Click on the "informations" tab in the application for more explanations. Here is a very quick summary...

5. Bayesian Analysis

Present the Bayesian Analysis applied in Chapter 7 in the thesis. This app also provides convenient tools to visualize the effects of changing the (hyper)parameters of the model(s), diagnostics, Predictive Posterior, ...

runExample('Bayesian') 

Click on the "informations" tab in the application for more explanations. Here is a very quick GIF summary : UPDATE : C++ backend for Gibbs sampling is now enabled, relying on the Rcpp R package. An $\approx$40 times increase of computational time efficiency is observed. Comparisons can be found inside the application for the generated chains.

All together in a dashboard

All applications created above in a smooth dashboard. Allows to put quite more tools.

runExample('All_dashboard') 

can be directly accessed by clicking on this link. However, it is recommended to visualize the Bayesian and Neural Networks applications individually since the dashboard's layout is not optimal with the panel divided in tabsets.

Where can you find the DATA ?

The project's data used for this thesis are confidential and have been provided by the Institut Royal de Météorologie from Uccle. Hence, I cannot put it as public and you must ask me if you want to obtain the data.

The data folder only contains the yearly data to allow use into the Shiny application.

proto4426/PissoortThesis documentation built on May 26, 2019, 10:31 a.m.