If this is used in the model without scaling, then the change in release_year from, for example, 1969 to 2021, could be interpreted as much more influential to the model results than the change from another variable if the difference there is only some number less than one. However, others like release_year are not. Many are on a scale of already, like acousticness, energy, danceability, and so on. In order to train many machine learning models, we need to scale the numeric variables. In order to do so, we will complete the steps below. Let us use the data to train some machine learning models. If I take a song from either my playlists or Brenda’s playlists, can I predict which set it was from using the data in the graph above? In other words, for each song, can the measures of its acousticness, energy, danceability, valence, liveness, speechiness, instrumentalness, loudness, length, tempo, popularity, mode, key, time_signature, and release_year be enough information to predict if it was on my playlists or on hers? Now that we have explored the data through visualization, let us shift gears to machine learning.

This shows the individual features for each song instead of an average from all 100. In the GIF below, features increase first by acousticness, then energy, danceability, and so on until release_year. She was listening to more new music than I was in 2016–2018, but I listened to more than her in 2021. My release_year is slightly lower in the first years but higher in the last one.Usually, songs are 4/4, but I have a few more songs that have 5/4, 6/4, or 7/4, and/or she has more that are lower like 3/4. This is just the average key (C, C#, D, etc.) so it makes sense that they are similar on average. I have more major songs than she does where she has more minor songs than I do in each playlist. I tend to like more popular songs than she does on average. My songs are a bit faster than hers on average. This is simply the average length of a song. This suggests, like energy, that my songs can be more noisy. Adding to the idea that I have more singing and she has slightly more instruments.

I have songs with more singing while she has slightly more instruments. I may have a couple more live versions of songs on my playlists. I like songs that are sad, depressing, and angry while she likes those that are more happy, cheerful, and euphoric. She likes music that is more suitable for dancing. My songs tend to be faster, louder, and noisier. She likes some more acoustic songs while I like more distortion. My acousticness is rather low, while Brenda’s is a bit higher.Here are some of the observations we can make from this graphic: My song features are plotted in darker colors while hers are plotted in a lighter overlay. Her playlists were added to help show more information of the plotted features based on our different tastes in music. The averages are taken across the 1,200 songs (600 from my playlists and 600 from Brenda’s playlists). One note is that each feature has been scaled to have a minimum of zero and a maximum of one. In the plot below, we graph the average of each feature taken of the 100 songs in each playlist for me and for my girlfriend as well. Notes: length is taken from duration_ms on page (1) and release_year is the year element of release_date on page (1): You can read more about them at these locations. These include the features: acousticness, energy, danceability, valence, liveness, speechiness, instrumentalness, loudness, length, tempo, popularity, mode, key, time_signature, and release_year for each song.

are dropped from the end of a word before plottingĪnother interesting visualization to consider contains musical features that are offered by Spotify’s API. have been removed since they are more common and do not provide as much meaning One of the largest words, feat, shows that a good amount of songs are featuring other artists. In addition, the word time in the bottom right shows that many songs are about feelings over time. For example, the word friend in the top left corner can show that some songs are about friendship. This offers an idea of the meaning of the songs. In this image, the size of the word represents how often it shows up. The total numbers accumulate as time goes on. The first graph below is a bar plot that shows my top 10 artists based on the number of their songs I had on my playlists in 2016-2021. In addition to looking at the songs per artist in a table, we can also look at it in a graph. And to see how many artists total were on each playlist: len_artists = len(df_2016].groupby('artist').count()) print(f'You had artists in the 100 songs in 2021.') You had 36 artists in the 100 songs in 2021.

Best visualizer for spotify 2018