In terminological/tidy-info-stats: Functions for manipulating information statistics

knitr::opts_chunk$set(echo = TRUE)
library(tidyinfostats)
library(tidyverse)
#devtools::load_all("..")
set.seed(101)

Tidy discretise

Purpose of tidy discretise is to generate discrete values in place in a tidy dataframe or database table

we'll use the iris dataset as an example

Manual data discretisation:

First generate a dataframe representation of the cuts we want. This is just a geometric combination of the cuts and the groups.

Sepal.length.cuts = c(5,5.5,7)
cutsDf = iris %>% group_by(Species) %>% cutsDfFromVector(Sepal.length.cuts)
cutsDf %>% group_by(Species) %>% standardPrintOutput::mergeCells()

Discretise the data:

#devtools::load_all("..")
#debug( discretise_Manual)
iris %>% group_by(Species) %>% discretise_Manual(Sepal.Length, Sepal.Length.Discrete, cutsDf, factorise=TRUE) %>% 
  group_by(Species,Sepal.Length.Discrete) %>% standardPrintOutput::mergeCells()

Bin count strategies:

Sometimes we want to vary the number of bins, or the location of the cuts on a group by group basis based on other parameters such as group size, or min and max values, or mean and sd of the grouped data

linearBySize can help create cuts based on the group count (n), min and max:

#devtools::load_all("..")
# slope defines the relationship between the number fo observations and the number of bins
binStrategy = linearBySize(slope=8,minBins=4,maxBins=100)

# e.g. if the sample had 16 measurements ranging from 0 to 128 the following function would generate the cutsDf
binStrategy(n=16,min=0,max=128)


binStrategy(n=64,min=0,max=20)

other bin strategies include:

• logNormalCentiles(bins=?)
• fixedNumber(bins=?)
• fixedWidth(width=?)

N.B. this is a WIP

Bringing this together

We can discretise data assuming a log normal distribution with $\mu$ = mean of the data, and $\sigma$ = var of the data.

N.B. this lognormal function is WIP - not working

• TODO: need to solve for mu and sigma....
• TODO: give user control of the labelling function...

#devtools::load_all("..")
#debug(discretise_ByValue)
iris %>% group_by(Species) %>% discretise_ByValue(Sepal.Length, Sepal.Length.Discrete, binStrategy = logNormalCentiles(bins=4)) %>%
  group_by(Species,Sepal.Length.Discrete) %>% 
  summarise(Count=n(), Sepal.Length.Mean = mean(Sepal.Length) ) %>%
  group_by(Species) %>%
  standardPrintOutput::mergeCells()

Discretisation by rank

E.g. if we want to get actual centiles in the data we can fix the number of bins. This will be quick but on the downside we don't get meaningful labels.

# devtools::load_all("..")
# undebug(discretise_ByRank)
iris %>% group_by(Species) %>% discretise_ByRank(Sepal.Length, Sepal.Length.Discrete, bins = 3) %>%
  group_by(Species,Sepal.Length.Discrete) %>% 
  summarise(Count=n(), Sepal.Length.Mean = mean(Sepal.Length) ) %>%
  group_by(Species) %>%
  standardPrintOutput::mergeCells()

terminological/tidy-info-stats documentation built on Nov. 19, 2022, 11:23 p.m.