.
1 - Introduccion
El siguiente trabajo consiste en utilizar los datos de todos los Pokemon para obtener insights respecto a sus cualidades, semejanzas y diferencias entre los mismos.
En primer lugar se va a realizar un análisis multivariado de los datos para luego pasar a un probar distintos modelos de reducción de la dimensionalidad para poder visualizar si efectivamente hay diferencia entre los Pokemon Legendarios y los No Legendarios.
2 - Analisis Multivariado
Librerías
library(tidyverse)
library(readxl)
library(highcharter)
library(flexdashboard)
library(Rtsne)Datos
df = read.csv('pokemon.csv')
txt = read_excel('values.xlsx', col_names = T)
dfgroup = df %>% group_by(type1, type2) %>% count()Total de Pokemons por Type
group_type = df %>% group_by(type1) %>% tally()
group_type %>%
arrange(desc(n)) %>%
hchart('column', hcaes(x = type1, y = n),
showInLegend = F,
maxSize = "15%",
dataLabels = list(enabled = TRUE,
format = '{point.y}')) %>%
hc_colorAxis(stops = color_stops(18, rev(RColorBrewer::brewer.pal(8, 'Dark2')))) %>%
hc_legend (enabled = FALSE) %>%
hc_tooltip(shared = FALSE,
borderColor = "black",
valueDecimals = 0) %>%
hc_add_theme(hc_theme_google()) %>%
hc_tooltip(pointFormat = '<b>Total: </b> {point.y} <br> ' ) %>%
hc_title(text = 'Type Pokemon',
style = list(fontSize = '20px', fontWeight = 'bold')) %>%
hc_subtitle(text = '<b>Bar Graph</b>',
style = list(fontSize = '16px', color = "black")) %>%
hc_credits(enabled = TRUE,
text = "Fuente: Kaggle",
align = "right",
verticalAlign = "bottom",
style = list(color = "black", fontSize = '15px'),
href = "https://www.kaggle.com/datasets/rounakbanik/pokemon") %>%
hc_yAxis(labels = list(style = list(color = "black")),
tickColor = "black") %>%
hc_xAxis(labels = list(style = list(color = "black")),
tickColor = "black") Top 20 Pokemonos sumando todas las habilidades:
df_bar = df[,c('hp','speed','attack','defense','type1','name', 'is_legendary', 'pokedex_number')]
df_bar$total = rowSums(df_bar[,1:4])
img_input = paste0("<img src ='", colnames(txt) , "shiny/{point.pokedex_number}.png' width='100' height='120'> <ul><li><b>Pokemon: {point.name}</li></b> <li><b>Hp: {point.hp} </b></li> <li><b>Attack: {point.attack} </b></li><li><b>Defense: {point.defense} </b></li><li><b>Speed: {point.speed} </b></li><li><b>Legendary: {point.is_legendary} </b></li><li><b>Pokedex: {point.pokedex_number} </b></li></ul> ")
get_bar_plot = function(data, n, filter, poketype) {
if (filter == TRUE){
data = data %>% filter(type1 == poketype)
}
data %>%
arrange(desc(total)) %>%
top_n(n) %>%
hchart('column', hcaes(x = name, y = total),
showInLegend = F,
maxSize = "15%",
dataLabels = list(enabled = TRUE,
format = '{point.y}')) %>%
hc_colorAxis(stops = color_stops(18, rev(RColorBrewer::brewer.pal(8, 'Set1')))) %>%
hc_legend (enabled = FALSE) %>%
hc_tooltip(pointFormat = img_input,
useHTML = T) %>%
hc_add_theme(hc_theme_google()) %>%
hc_title(text = 'Type Pokemon',
style = list(fontSize = '20px',
fontWeight = 'bold')) %>%
hc_subtitle(text = '<b>Bar Graph</b>',
style = list(fontSize = '16px',
color = "black")) %>%
hc_credits(enabled = TRUE,
text = "Fuente: Kaggle",
align = "right",
verticalAlign = "bottom",
style = list(color = "black",
fontSize = '15px'),
href = "https://www.kaggle.com/datasets/rounakbanik/pokemon") %>%
hc_yAxis(labels = list(style = list(color = "black")),
tickColor = "black") %>%
hc_xAxis(labels = list(style = list(color = "black")),
tickColor = "black")
}
get_bar_plot(df_bar, 20, FALSE)Ultimos 20 Pokemons sumando todas las habilidades:
get_bar_plot(df_bar, -20, FALSE)## Selecting by totalAhora veamos los tipo Electric:
get_bar_plot(df_bar, 10, TRUE, 'electric')## Selecting by totalEn el siguiente gráfico podemos visualizar un Scatter Plor con la relación entre el nivel de ataque y defensa de todos los Pokemons agrupados por tipo.
Si dejamos el cursor sobre cada uno de los puntos del Scatter podemos ver todos los niveles de velocidad, ataque y defensa de cada uno de ellos.
df %>%
hchart("scatter", hcaes(x = attack,
y = defense,
name = name,
group = type1),
maxSize = "5%",
regression = FALSE,
regressionSettings = list(
type = "polynomial",
color = 'steelblue',
dashStyle = "ShortDash",
tooltip = FALSE,
decimalPlaces = 6,
useAllSerues = TRUE,
order = 1,
lineWidth = 5,
name = "%eq | r2: %r",
hideInLegend = TRUE)) %>%
hc_tooltip(pointFormat = img_input,
useHTML = T) %>%
hc_add_dependency("plugins/highcharts-regression.js") %>%
hc_add_theme(hc_theme_google()) %>%
hc_credits(enabled = TRUE, text = "Fuente: Kaggle",align = "right",verticalAlign = "bottom",
style = list(color = "black", fontSize = '15px'),
href = "https://www.kaggle.com/datasets/rounakbanik/pokemon") %>%
hc_title(text = paste0('Attack vs Defense'),
style = list(fontSize = '20px', fontWeight = 'bold'))df %>%
hchart("scatter", hcaes(x = speed,
y = weight_kg,
name = name,
group = type1),
maxSize = "5%",
regression = FALSE,
regressionSettings = list(
type = "polynomial",
color = 'steelblue',
dashStyle = "ShortDash",
tooltip = FALSE,
decimalPlaces = 6,
useAllSerues = TRUE,
order = 1,
lineWidth = 5,
name = "%eq | r2: %r",
hideInLegend = TRUE)) %>%
hc_tooltip(pointFormat = img_input,
useHTML = T) %>%
hc_add_dependency("plugins/highcharts-regression.js") %>%
hc_add_theme(hc_theme_google()) %>%
hc_credits(enabled = TRUE, text = "Fuente: Kaggle",align = "right",verticalAlign = "bottom",
style = list(color = "black", fontSize = '15px'),
href = "https://www.kaggle.com/datasets/rounakbanik/pokemon") %>%
hc_title(text = paste0('Weight vs Speed'),
style = list(fontSize = '20px', fontWeight = 'bold'))dat <- data_to_boxplot(df,
attack,
type1,
name = "height in meters",
add_outliers = FALSE,
color = "black")
highchart() %>%
hc_xAxis(type = "category" ) %>%
hc_add_series_list(dat) %>%
hc_add_series(
data = df,
type = "scatter",
marker = list(radius = 2.5), alpha = 0.9,
hcaes(x = type1, y = attack, group = type1)
) %>%
hc_plotOptions(boxplot = list(
fillColor = '#F0F0E0',
lineWidth = 2,
medianColor = '#0C5DA5',
medianWidth = 3,
stemColor = '#A63400',
stemDashStyle = 'dot',
stemWidth = 1,
whiskerColor = '#3D9200',
whiskerLength = '20%',
whiskerWidth = 3,
color = 'black'
)) %>%
hc_legend (enabled = FALSE) %>%
hc_title(text = 'Box Plot Pokemon',
style = list(fontSize = '20px',
fontWeight = 'bold')) %>%
hc_subtitle(text = paste0('<b>','attack','</b>'),
style = list(fontSize = '16px',
color = "black")) %>%
hc_credits(enabled = TRUE,
text = "Fuente: Kaggle",
align = "right",
verticalAlign = "bottom",
style = list(color = "black",
fontSize = '15px'),
href = "https://www.kaggle.com/datasets/rounakbanik/pokemon") %>%
hc_yAxis(labels = list(style = list(color = "black")),
tickColor = "black",
title = list(text = paste0('<b>','attack','</b>'))) %>%
hc_xAxis(labels = list(style = list(color = "black")),
tickColor = "black") 3 - Algoritmos
PCA:
PCA es uno de los métodos más importantes de reducción de dimensionalidad para visualizar datos. PCA es una técnica que convierte n-dimensiones de datos en k-dimensiones mientras mantiene la mayor cantidad de información del conjunto de datos original
TSNE:
El algoritmo t-SNE consiste en crear una distribución de probabilidad que represente las similitudes entre vecinos en un espacio de gran dimensión y en un espacio de menor dimensión. Por similitud, intentaremos convertir las distancias en probabilidades
Diferencias entre PCA y T-SNE:
PCA es una técnica de reducción de dimensionalidad lineal mientras que T-SNE no lo es. PCA intenta preservar la estructura global de los datos mientras que T-SNE preserva la estructura local. PCA se ve muy afectado por los valores atípicos, mientras que T-SNE puede manejar los valores atípicos.
Podemos observar que los algoritmos de reducción de la dimensionalidad nos dan información sobre la diferencia entre los Pokemons legendarios y no legendarios. Para este caso pasamos de 6 dimensiones a solo 2.
set.seed(15)
df_model = df[,c('sp_attack','sp_defense','hp','speed','attack','defense','type1','name', 'is_legendary', 'pokedex_number')]
tsne = Rtsne(df_model[,c('sp_attack','sp_defense','hp','speed','attack','defense')],
perplexity = 30,
eta = 1000,
max_iter = 5000,
check_duplicates = FALSE)
Y = as.data.frame(tsne$Y)
df_model$X = Y$V1
df_model$Y = Y$V2
df_model %>%
hchart("scatter", hcaes(x = df_model$X,
y = df_model$Y,
name = name,
group = is_legendary),
maxSize = "5%",
regression = FALSE,
regressionSettings = list(
type = "polynomial",
color = 'steelblue',
dashStyle = "ShortDash",
order = 1,
lineWidth = 5,
name = "%eq | r2: %r",
hideInLegend = TRUE)) %>%
hc_tooltip(pointFormat = img_input,
useHTML = T) %>%
hc_add_theme(hc_theme_google()) %>%
hc_add_dependency("plugins/highcharts-regression.js") %>%
hc_credits(enabled = TRUE, text = "Fuente: Kaggle",align = "right",verticalAlign = "bottom",
style = list(color = "black", fontSize = '15px'),
href = "https://www.kaggle.com/datasets/rounakbanik/pokemon") %>%
hc_title(text = paste0('Pokemon RTSNE Model - Legendary'),
style = list(fontSize = '20px', fontWeight = 'bold'))PCA <- princomp(df_model[,c('sp_attack','sp_defense','hp','speed','attack','defense')],
cor = T)
df_scores = as.data.frame(PCA$scores)
df_model$X_pca = df_scores$Comp.1
df_model$Y_pca = df_scores$Comp.2
df_model %>%
hchart("scatter", hcaes(x = df_model$X_pca,
y = df_model$Y_pca,
name = name,
group = is_legendary),
maxSize = "5%",
regression = FALSE,
regressionSettings = list(
type = "polynomial",
color = 'steelblue',
dashStyle = "ShortDash",
order = 1,
lineWidth = 5,
name = "%eq | r2: %r",
hideInLegend = TRUE)) %>%
hc_tooltip(pointFormat = img_input,
useHTML = T) %>%
hc_colors(c('#43c7d9', '#9c38c7')) %>%
hc_add_theme(hc_theme_google()) %>%
hc_add_dependency("plugins/highcharts-regression.js") %>%
hc_credits(enabled = TRUE, text = "Fuente: Kaggle",align = "right",verticalAlign = "bottom",
style = list(color = "black", fontSize = '15px'),
href = "https://www.kaggle.com/datasets/rounakbanik/pokemon") %>%
hc_title(text = paste0('Pokemon PCA Model - Legendary'),
style = list(fontSize = '20px', fontWeight = 'bold'))