統計學習與分析期末專題

Load library

library(dplyr)
library(data.table)
library(ggplot2)
library(treemap)
library(googleVis)
library(flexclust)
library(FactoMineR)
library(factoextra)

Load Data

songs=fread('../data/songs.csv')
songs

Artist_top10.table=songs %>% 
  filter(.,Top10==1) %>%
  group_by(.,artistname) %>%
  summarise(
    num=n()
  )
Artist_top10.table

Artist_top10.table[order(Artist_top10.table$num,decreasing = T),] %>% 
  head(10) %>% 
  ggplot(data=.) +
  geom_bar(aes(x=substr(artistname,start=0,stop=5),y=num,fill=artistname),stat = "identity")+
  xlab('ArtistName')+
  ylab('#Top10 songs ')

Analysis the most Keys

treemap

Key=c('C','#C','D','#D','E','F','#F','G','#G','A','#A','B')
songs$Key_str=Key[songs$key+1]

mostkey.data=songs %>% 
  group_by(.,Key_str) %>%
  summarise(
    num=n()
  )
mostkey.data

treemap(mostkey.data,
        index='Key_str',
        vSize = 'num',
        type='value',
        vColor = 'num', 
        palette = rainbow(6),
        title='#Key'
)

Analysis the most tempo types

x <- songs$tempo
songs$tempo_type=case_when(
  x >=66 & x<76 ~ "Adagio",
  x >=76 & x<108 ~ "Andante",
  x >=108 & x<120 ~ "Moderato",
  x >=120 & x<156 ~ "Allegro",
  x >=156 & x<176 ~ "Vivace",
  x >=176  ~ "Presto",
  TRUE ~"Others"
)
songs$tlabel=case_when(
  x >=66 & x<76 ~ "66-76",
  x >=76 & x<108 ~ "76-108",
  x >=108 & x<120 ~ "108-120",
  x >=120 & x<156 ~ "120-156",
  x >=156 & x<176 ~ "156-176",
  x >=176  ~ ">176",
  TRUE ~"< 66 BPM"
)

most_temptype.table=songs %>% 
  group_by(.,tempo_type,tlabel) %>%
  summarise(
    num=n()
    
  )
most_temptype.table

most_temptype.table %>%
ggplot(data=., aes(x=reorder(tempo_type,num), y=num))+
  geom_bar(aes(y=num),stat="identity", alpha=0.8,fill="skyblue" )+ 
  ylab("Count")+ xlab("Tempo Type")+
  ggtitle("What is the most popular Tempo type? ")+
  geom_text(aes(label=tlabel), vjust=1, color="maroon", size=3.5)+ 
  theme_minimal()

Analysis Temple、Key、Top10

temp_key_top10.table= songs %>%
  filter(Top10==1) %>%
  group_by(.,tempo_type,Key_str) %>%
  summarise(
    num=n()
  )
temp_key_top10.table

Heat map

temp_key_top10.table %>%
ggplot(., aes(x=tempo_type, y=Key_str, fill = num)) + 
    geom_tile(colour = "white")  +
    scale_fill_gradient(low="skyblue", high="Pink") +
    labs(x="Tempo", y=NULL, title="Heatmap of #Top10" ,fill="#Top10")

分群分析:

#dist.method:Euclidean
#hclust.method:Complete
songs[,c(8,9,11:13)]%>% names()

[1] "loudness"       "tempo"          "key"            "key_confidence" "energy"        

hc = songs[,c(8,9,11:13)] %>% scale %>% dist %>% hclust
plot(hc,labels=FALSE)
rect.hclust(hc, k=3, border="red")

開啟平行運算

# library(doParallel)
# clust = makeCluster(detectCores())
# registerDoParallel(clust); getDoParWorkers()

層級式分群視覺化

library(FactoMineR)
library(factoextra)
fviz_dend(
  hc, k=3, show_labels=F, rect=T, rect_fill=T,
  labels_track_height=0,
  palette="ucscgb", rect_border="ucscgb")

視覺化分群

songs$group = cutree(hc, k=3) %>% factor
ggplot(songs, aes(x=tempo, y=loudness, col=group)) +
  geom_point(size=3, alpha=0.5) + 
  theme_light()

PCA降維視覺化

songs[,c(8,9,11:13)] %>% PCA(graph=F) %>% fviz_pca_biplot(
  label="var", col.ind=songs$group,
  pointshape=19, mean.point=F,
  addEllipses=T, ellipse.level=0.7,
  ellipse.type = "convex", palette="ucscgb",
  repel=T
  )

切割群組

grp = cutree(hc, k = 3)
table(grp)

grp
   1    2    3 
4706 2725  143 

群組屬性

sapply(split(songs[,c(8,9,13:15)], grp), colMeans) %>% round(3)

                  1       2       3
loudness     -8.206  -9.290 -19.908
tempo        94.588 130.682  82.628
energy        0.687   0.680   0.192
pitch         0.010   0.011   0.014
timbre_0_min  4.048   4.272   3.720

資料視覺化

table(grp)

grp
   1    2    3 
4706 2725  143 

layout(matrix(c(1,2,2), 3, 1))
par(mar=c(2,3,1,1), cex=0.8)
table(grp) %>% barplot(col=3:5, names.arg=paste0("Group-",1:3))
par(mar=c(6,3,2,1))

sapply(split(songs[,c(8,9,13:15)], grp), colMeans) %>% t %>% 
  barplot(beside=T, col=3:5, las=2)