###
###  ELENA/KEEL Phoneme Data
###

Local copy:
  phoneme.csv

Downloaded From:
  https://www.openml.org/d/1489
  (CSV file: https://api.openml.org/data/v1/get_csv/1592281)

Original Source:
  ELENA project (European ESPRIT 5516 project ROARS), 1993; formatted by the KEEL repository
  https://www.elen.ucl.ac.be/neural-nets/Research/Projects/ELENA/databases/REAL/phoneme/
  https://sci2s.ugr.es/keel/dataset.php?cod=105

Variable Information:

  Name -- Data Type -- Measurement -- Description
  V1    -- quantitative -- standardized, unitless -- amplitude of the 1st harmonic, AH1/Ene
  V2    -- quantitative -- standardized, unitless -- amplitude of the 2nd harmonic, AH2/Ene
  V3    -- quantitative -- standardized, unitless -- amplitude of the 3rd harmonic, AH3/Ene
  V4    -- quantitative -- standardized, unitless -- amplitude of the 4th harmonic, AH4/Ene
  V5    -- quantitative -- standardized, unitless -- amplitude of the 5th harmonic, AH5/Ene
  Class -- binary (1, 2) -- none -- 1 = nasal vowel, 2 = oral vowel -- response variable

Summary:
  Number of instances (observations): 5404
  Number of Attributes: 6
  Attribute breakdown: 5 quantitative feature variables, and 1 binary response variable
  Missing Attribute Values: None
  Class distribution: class 1 (nasal) = 3818 (70.7%), class 2 (oral) = 1586 (29.3%)


Abstract:
  The aim is to distinguish between nasal and oral vowels using five acoustic
  measurements of each vowel. The data comes from the ROARS project,
  which developed a real-time analytical speech recognition system for French and Spanish.

Data Characteristics:
  Vowels come from 1809 isolated syllables (for example: pa, ta, pan, ...).
  Three observation moments were kept for each vowel: the observation at maximum
  total energy (Ene), and the observations 8 msec before and 8 msec after it.
  This gave 5427 instances; 23 instances whose five first harmonics all had
  amplitude zero were removed, leaving 5404. Patterns are in random order.
  The attributes in the OpenML copy are standardized (mean 0, standard deviation 1).

Notes:
  The data was created by Dominique Van Cappel, THOMSON-SINTRA.
  AHi/Ene is the amplitude of harmonic i normalised by the total energy Ene
  (integrated over all frequencies). Each harmonic is signed: positive when it
  corresponds to a local maximum of the spectrum and negative otherwise.
  The original ELENA documentation codes the classes as 0 = nasal and 1 = oral;
  the OpenML file shows them as 1 and 2 (mapping inferred from the class sizes,
  3818 nasal and 1586 oral).


Reference:
  Alinat, P., Periodic Progress Report 4, ROARS Project ESPRIT II - Number 5516,
  February 1993, Thomson report TS. ASM 93/S/EGS/NC/079.
  Guerin-Dugue, A. and others, Deliverable R3-B4-P - Task B4: Benchmarks, Technical
  report, Elena-NervesII "Enhanced Learning for Evolutive Neural Architecture",
  ESPRIT-Basic Research Project Number 6891, June 1995.
  Verleysen, M. and Voz, J.L. and Thissen, P. and Legat, J.D., A statistical Neural
  Network for high-dimensional vector classification, ICNN'95 - IEEE International
  Conference on Neural Networks, November 1995, Perth, Western Australia.


----------
library(tidyverse)
DF1 <- read_csv("https://nmimoto.github.io/datasets/phoneme.csv")
DF1


# Define function that converts (1,2) to (No, Yes)
thisFunc01 = function(x) {
    case_when(x == 1 ~ "No",
              x == 2 ~ "Yes",
              .default = NA)}

# Apply the function
DF2 <- DF1 %>% mutate_at("Class", thisFunc01) %>%
    mutate_at("Class", as.factor) %>%
    mutate(Class = factor(Class, levels=c("No", "Yes")))

DF2
levels(DF2$Class)   # Important that the order is ("No", "Yes")