np-completed / group_lasso

Statistical regularization

group lasso

The group_lasso project performs regularization using the Sonar dataset.

Background

This README file shows the technical implementation for statistical regularization. All discussion, results and background are on display as PDF or HTML docs, associated with a .Rmd file from R.

• Setup the WIKI page of GitHub for this project.

• Remove missing data command Sonar_completeCase = Sonar[complete.cases(Sonar),] , bc there is no missing data.

• Migrate to sparklyr for utilizing H2O.

• Group YML file for parent-child docs

GitHub Setup:

Create new repository from command line

echo "# group_lasso" >> README.md
git init
git add README.md
git commit -m "first commit"
git remote add origin https://github.com/prfrl/group_lasso.git
git push -u origin master

Project Directories

THe GitHub directories follow my local project directories. See directory_structure.png to see a picture.

The directory tree is logically formatted, using best practices (maybe). All files are placed relative to the root, where the .gitignore file is located.

• code
  • deep_learn_h2o_roxy.R Deep Learning Machine learning models run in H2O.File uses roxygen formatting.
  • deep_learn_h2o_plain.R A non-roxygen version kept as backup, just in case.
  • helper_h2o.R
  • main.Rmd Main File Parent (main) file aggregates child R files.
  • main.html
  • main_cache/html Cache files are ignored to save space on GitHub.
  • main_files/figure-html Many plots are ignored, to save space.

requirements

R libraries

The knitr md file group_lasso generated the HTML output using R libraries: group_lasso/code/grp_lasso_cache/html/__packages . All bleeding edge packages also have install directions shown in the .Rmd file. Typically library(devtools) and install_github(foo_bar/magick) .

Roxygen

The main file main.Rmd imports all R{MD} script(s) and generates a unified output. The child files can be buggy, predictably tables and charts, but there are many benefits to logical compartmentalizing scripts.

Roxygen formatting

R files formatted using roxygen can be rendered using pandoc, command line arguments using a bash shell OR R. Most markdown effects persist, however the main two changes are:

1. Plain texts in roxygen comments #' are preserved as normal texts (may contain inline R code)

2. chunk options are written after #+ or #-, e.g. #+ chunk-label, opt1=value1, opt2=value2

There are three main ways to run roxygen: https://cran.r-project.org/web/packages/roxygen2/vignettes/roxygen2.html

1. roxygen2::roxygenise()
2. devtools::document() , if you’re using devtools
3. Ctrl + Shift + D, if you’re using RStudio.

Ideally a main YML file is used and pulled from all existing parent child files. This has not happened yet.

Parent File Output from Multiple Child Files

Simple solution is to run source(foo_bar.R) from the parent file. this is simple but not flexible. Output of tables and plots can be formatted using HTML from the plot figures. Variables are callable without too much fuss.

Compiling repors from R scripts: https://rmarkdown.rstudio.com/articles_report_from_r_script.html