How Likely is Each Genre?

Do the genres your hear played on the shuffle respresent the genres of the song in our class playlist?

As a class, we put together a playlist of 144 of our favorite songs, and want to know whether they will be played a equal amount proportionate to the amount of songs in each recongnized genre.

sampfull<-df %>% count(genre)%>% mutate(theoretic_probability = n/sum(n))
sampfull
## # A tibble: 6 × 3
##   genre                        n theoretic_probability
##   <chr>                    <int>                 <dbl>
## 1 Genre 1 - hiphop/rap        26                0.181 
## 2 Genre 2 - pop/kpop/Latin    22                0.153 
## 3 Genre 3 -  house/ska         5                0.0347
## 4 Genre 4 -  rnb/soul         43                0.299 
## 5 Genre 5 - alt/indie/folk    34                0.236 
## 6 Genre 6 - country/rock      14                0.0972

Theoretical Proabbility of each genre being played

With 144 songs on the playlist, this is the theoretical percentage of each genre being “played”:

- Hiphop/Rap: 18%, pop/kpop/Latin: 15%, House/Ska: 4%, rnb/soul: 30%, - alt/indie/folk: 24%, country/rock: 10%.

When you play shuffle on spotify, it “randomly” selects a song from this playlist to play. The values above show how likely it is for each of our genres to play if you click shuffle.

df5<- sample_n(df, 10)
samp10<-df5 %>% count(genre)%>% mutate(Experimental_probability = n/sum(n))
samp10
## # A tibble: 5 × 3
##   genre                        n Experimental_probability
##   <chr>                    <int>                    <dbl>
## 1 Genre 1 - hiphop/rap         1                      0.1
## 2 Genre 2 - pop/kpop/Latin     2                      0.2
## 3 Genre 4 -  rnb/soul          3                      0.3
## 4 Genre 5 - alt/indie/folk     3                      0.3
## 5 Genre 6 - country/rock       1                      0.1

After Randomly generating 10 songs from our playlist, these were the choices.

In this 10 song sample size, rnb/soul came up 3 times, pop/kpop/Latin came up 4 times, country/rock was pulled twice, and hiphop/rap was pulled once. Two genres were not generated in this simulation, both house/ska, and indie/alt/folk.

With this data, we can calculate the experimental probability of each genre picking played on shuffle

- rnb/soul 30%, alt/indie/folk 0%, pop/kpop/Latin 40%, house/ska 0%, rock/country 20%, hiphop/rap 10%

Is this a good representaion of songs in our playlist?

I believe its too small a sample size to properally represent our whole playlist. Pulling these 10 songs, we got no house/ska and no indie/alt/folk, so it is difficult to say this is a proper representation of our full playlist.

What about other sample sizes?

df2<- sample_n(df, 15)
df2
## # A tibble: 15 × 3
##    genre                    artist              title        
##    <chr>                    <chr>               <chr>        
##  1 Genre 4 -  rnb/soul      Amerie              Can't Let Go 
##  2 Genre 5 - alt/indie/folk Death Cab for Cutie No Sunlight  
##  3 Genre 1 - hiphop/rap     PnB Rock            Selfish      
##  4 Genre 2 - pop/kpop/Latin Taeyang             Wedding dress
##  5 Genre 5 - alt/indie/folk Sorry               Showgirl     
##  6 Genre 1 - hiphop/rap     Russ                Handsomer    
##  7 Genre 1 - hiphop/rap     Common              Be           
##  8 Genre 4 -  rnb/soul      sza                 normal girl  
##  9 Genre 4 -  rnb/soul      Beyonce             Summertime   
## 10 Genre 4 -  rnb/soul      Thee sacred souls   For now      
## 11 Genre 6 - country/rock   Citizen             Hyper Trophy 
## 12 Genre 1 - hiphop/rap     Ken carson          Murda musik  
## 13 Genre 2 - pop/kpop/Latin Post Malone         Goodbyes     
## 14 Genre 2 - pop/kpop/Latin renee rap           tummy hurts  
## 15 Genre 5 - alt/indie/folk Taylor Swift        Peace
samp15<-df2 %>% count(genre)%>% mutate(Experimental_probability = n/sum(n))
samp15
## # A tibble: 5 × 3
##   genre                        n Experimental_probability
##   <chr>                    <int>                    <dbl>
## 1 Genre 1 - hiphop/rap         4                   0.267 
## 2 Genre 2 - pop/kpop/Latin     3                   0.2   
## 3 Genre 4 -  rnb/soul          4                   0.267 
## 4 Genre 5 - alt/indie/folk     3                   0.2   
## 5 Genre 6 - country/rock       1                   0.0667

This sample pulled 15 songs and our yeilded a different experimental probability for each genre:

- Genre 1 - hiphop/rap 20%, Genre 2 - pop/kpop/Latin 6.6%, Genre 3 - house/ska 6.6%, Genre 4 - rnb/soul 33.3%, Genre 5 - alt/indie/folk, 26.6%, and Genre 6 - country/rock 6.6%.

This slightly larger sample size seemed to yeild a more similar probability to our previously calculated theorhectical probability.

What if the sample size is smaller?

df3<- sample_n(df, 5)
df3
## # A tibble: 5 × 3
##   genre                artist       title                     
##   <chr>                <chr>        <chr>                     
## 1 Genre 4 -  rnb/soul  Beyonce      Summertime                
## 2 Genre 4 -  rnb/soul  Leon Bridges River                     
## 3 Genre 4 -  rnb/soul  Sade         Keep Looking              
## 4 Genre 4 -  rnb/soul  Marvin Gaye  Just to keep you satisfied
## 5 Genre 1 - hiphop/rap Kanye west   Missed call
samp5<-df3 %>% count(genre)%>% mutate(Experimental_probability = n/sum(n))
samp5
## # A tibble: 2 × 3
##   genre                    n Experimental_probability
##   <chr>                <int>                    <dbl>
## 1 Genre 1 - hiphop/rap     1                      0.2
## 2 Genre 4 -  rnb/soul      4                      0.8

This smaller sample size is practically doomed from the start. Simulating only five playlist shuffles, it will never be able to include all six of our genres to accurately represent them. In reality it only simulated four genres!

-Genre 2 - pop/kpop/Latin 20%, Genre 4 - rnb/soul 40%, Genre 5 - alt/indie/folk 20%, Genre 6 - country/rock 20%.

In this sample size, two genres were not represented, both House/Ska, and hiphop/rap. The experimental probability of these genres were very different from out original theorhetical values.

All of our sample sizes so far were fairly small, and didnt represent the original set of data closely, to test if the sample size makes a difference on the correlation, lets try a much bigger sample size.

df4<- sample_n(df, 50)
df4
## # A tibble: 50 × 3
##    genre                    artist       title                  
##    <chr>                    <chr>        <chr>                  
##  1 Genre 5 - alt/indie/folk Ginger Root  Loretta                
##  2 Genre 2 - pop/kpop/Latin renee rap    tummy hurts            
##  3 Genre 1 - hiphop/rap     Stormzy      Hide & Seek (FLO Remix)
##  4 Genre 4 -  rnb/soul      Lauryn Hill  Tell Him               
##  5 Genre 5 - alt/indie/folk Thundercat   Them Changes           
##  6 Genre 4 -  rnb/soul      Solange      Don't Touch My Hair    
##  7 Genre 1 - hiphop/rap     Lakeyah      Worst thing            
##  8 Genre 6 - country/rock   Dolly parton Jolene                 
##  9 Genre 1 - hiphop/rap     Kanye West   Everything I Am        
## 10 Genre 1 - hiphop/rap     Mac Miller   Surf                   
## # ℹ 40 more rows
samp50<-df4 %>% count(genre)%>% mutate(Experimental_probability = n/sum(n))
samp50
## # A tibble: 6 × 3
##   genre                        n Experimental_probability
##   <chr>                    <int>                    <dbl>
## 1 Genre 1 - hiphop/rap        12                     0.24
## 2 Genre 2 - pop/kpop/Latin    10                     0.2 
## 3 Genre 3 -  house/ska         2                     0.04
## 4 Genre 4 -  rnb/soul         14                     0.28
## 5 Genre 5 - alt/indie/folk     9                     0.18
## 6 Genre 6 - country/rock       3                     0.06

This sample size pulled 50 songs, around a third of the original data. The experimental probablities were:

- Genre 1 - hiphop/rap 18%, Genre 2 - pop/kpop/Latin 16%, Genre 4 - rnb/soul 26%, Genre 5 - alt/indie/folk 26%, Genre 6 - country/rock 14%

These values show very close correlation with our original probabilities, although there still is a genre unrepresented: House/Ska. This genre only had 5 songs in the whole sample size, so it was a fairly low chance of it getting selected, but this can still show how incorrectly sized samples cannot fairly represent data.

sampfull<-sampfull%>%select (genre, theoretic_probability)
samp50<-samp50%>%select (genre, Experimental_probability)
fulltable<-full_join(sampfull, samp50, by = "genre")
fulltable
## # A tibble: 6 × 3
##   genre                    theoretic_probability Experimental_probability
##   <chr>                                    <dbl>                    <dbl>
## 1 Genre 1 - hiphop/rap                    0.181                      0.24
## 2 Genre 2 - pop/kpop/Latin                0.153                      0.2 
## 3 Genre 3 -  house/ska                    0.0347                     0.04
## 4 Genre 4 -  rnb/soul                     0.299                      0.28
## 5 Genre 5 - alt/indie/folk                0.236                      0.18
## 6 Genre 6 - country/rock                  0.0972                     0.06

This final model ties in both our original theorhetical probability calculated in the begining, as well the experimental probability for our 50 song experimental sample. Using this, it is easier to compare the two data sets.

Analysis and Reflection:

1. Analysis

- To decide how likely each genre getting played really is, you have to pick the sample size you believe fits the data the best.

- The larger the sample size, the more accuratte data it will provide, so to discuss the the experimental probability, i will focus on my 50 song sample group.

- This sample is extremly similar to the therorhetical probabiliy origanlly calculated. The differences between the two samples is negligable and if resimulated could show slightly different results, all due to chance.

- This project has everything to do with data science, as we used science to analyze our data! We collected our data and cleaned it, analyzed it by building a model to represent what we have gathered, and presented it in this very file.

2. Reflection

- The process of coding this project allowed me to view the original data at the right angle to answer my question I posed at the beginning of my research, and conclude that the songs played on shuffle reflect the statistics fairly well, but not often exactly.

- The most challenging code I wrote throughout the project was not specifically one section, but just remembering to name all my data, models, and specific sections. Many times i had to go back and alter all my code because of incorrect labelling, or somtimes none at all.

- I feel most proud of the last chunk of code because of the way it is able to synthesize my previous models and data.

- I really love the formatting aspect of coding; seeing how all chunks fit together in a consise way, and playing around with code until it looks neat and tidy.

- If i were to do this project again i would like to play around with the theme of the knit file, possible coloring and models I made, or finding new ways to format my text.