Closed jgomezm closed 4 years ago
@rgurlek Should we have more countries and make it multiclass?
We can. Will it make the story more appealing?
Adding more classes could help form a fun data story if this is our algorithm. If we don't use more classes, we have what we have. We can improve iteratively by changing parameters and architectures. If we add more classes, we can look at the confusion matrix and try to tell a story around it. Are countries that get confused similar in any identifiable dimension such as language, proximity (same continent), or culture (anglo-saxon vs. middle eastern vs. east asian)?
Stepping away from countries, we can cluster by year of release aggregating across countries. Similarly, we could see if years in close proximity to one another get confused more
You are right! Having clusters of countries can make it interesting.
Ok, I'm going to collect all the countries Spotify is active in over several years to get a bigger n. I'll probably drop month
Collecting all countries listed at the link below for playlists on June 15th from 2000 to 2019
Pick a set of differentiable playlists