In erblast/oetteR: Collection of personal R functions

knitr::opts_chunk$set(echo = TRUE, message = FALSE, warning = FALSE )

suppressPackageStartupMessages( require(oetteR) ) 
suppressPackageStartupMessages( require(tidyverse) )

Cleaning Data

Before I learned about recipes. I had my own set of functions that would do something similar with less functionality of course. oetteR::f_clean_data() takes a dataframe and performs some automated cleaning steps and sorts the variables into categrocial and numerical categories and returns a list which I always name data_ls.

data = ISLR::Auto

data_ls = f_clean_data( data = data
                         # reduce number of levels to 10, group to other
                         , max_number_of_levels_factors = 10
                         # numericals with less than 10 unique values will be converted to factors
                         , min_number_of_levels_nums = 10
                         # exclude missing data
                         , exclude_missing = T
                         # negative values will be set to zero
                         ,replace_neg_values_with_zero = T
                         # allow negative values in these columns
                         ,allow_neg_values = 'null'
                         # tag id columns
                         , id_cols = 'name'
                         )


print( str(data_ls) )

Notice that numericals are converted to ordered factors

BoxCox Transformation

We can use a boxcox transformation on numerical variables.

data_ls = f_boxcox(data_ls)

print( str(data_ls) )

PCA

Base R has a decent function for PCA prcomp(). However the returned object is a bit messy and contains a lot of matrices intesad of dataframes. oetteR::f_pca() is a convenience wrapper which uses data_ls lists.

pca_ls = f_pca( data_ls
             , center = T
             , scale = T
             , use_boxcox_tansformed_vars = T
             , include_ordered_categoricals = T
             )

The returned pca object has some new features compared to prcomp()

Call the original Data with the Principle Components as additional Variables

as_tibble( pca_ls$data )

Get variance explained as a dataframe

pca_ls$pca$vae

How much is each variable contributing to each component in percent

Note the columns add up to 100

pca_ls$pca$contrib

Contribution of variables to variance explained by principle component

pca_ls$pca$contrib_abs_perc

Plot PCA

Plot variance explained

Note returns a plotly graph by default

oetteR::f_pca_plot_variance_explained

f_pca_plot_variance_explained(pca_ls
                              # dont include componetns that explain less than 2.5 percent of the variabnce
                              , threshold_vae_for_pc_perc = 2.5)

Plot first two principle components and color by cylinder

and color cylinders, oetteR::f_pca_plot_variance_explained returns a taglist created by htmltools::tagList which stores a plotly graph and a DT:datatable. The algebraic sign (+/-) of the rotation value in the table tells you whether a contribution of one variable to one principle component is positive or negative.

taglist = f_pca_plot_components( pca_ls
                                 , x_axis = 'PC1'
                                 , y_axis = 'PC2'
                                 , group = 'cylinders' )

taglist