michaelmoschitto / aisongrecommender Goto Github PK

Do this

Although not part of the first sprint we have a goal of being able to classify songs based on genre. This will require manually compiling songs for different genres

Research and select a UI package to use for front end development to ensure a clean interface.

If we are going to be training and testing on user specific songs were are going to want to be able to store the songs that we scrape for a given user into a DB so as to aboid throttling

Find layers that will aid in classification

We need a list of main genre's to be able to say that an artist has belongs to this genre.

Ex: drake has canadian rap, toronto rap, rap, but we just want him to be in the genre rap for our purposes.

How are we goign to convey our results

Using preferences and user data, construct a large dataset based on their music tastes.

Hyperparemeter tuning to best feasible accuracy

Should start with a baseline accuracy from Random Forest, Bayes, Knn, Kmeans

Available libraries / easiest way to do this

There are multiple ways to get data from spotify and we need to figure out if the web API is the same as the information from the Spotipy python lib

Notebook only, notebook where source code is in Python file, Python only.

If notebook, where would it be hosted?

hopefully using the right features will increase model accuracy

Related to metrics research but we need to know which things the API will allow

Find correlations between features of the song dataset

Find out how to pull song data given user input

Using artist compile a list of the most popular gengre for a users playllist.

We are going to have to use this method to if we want to predict a song from a certain genre becuase the spotify API doesn't expose genre yet.

Find an example of a neural net that will beat ours

Python -> MySQL lib

Rough draft of classifier not worrying about accuracy

Once we have the data formatted and know they type, we need to decide on which classifier is the best

Set up a local React website and confirm all members can successfully run the website.

Take notes on our first meeting

Create a mock-up for the website to establish functionality goals and a target interface.

Regression: which kind, NN types of layers, activtion, KNN, Kmeans

Set up CORS and Axios to allow for communication between website and backend code. This will allow the website to retrieve the information processed by the data model.

lots of different metrics on spotify related to songs, users, playlists, etc. We should have a list of all of them or at the very least which ones we want to target.

Part of our initial data exploration
Graph avg of each song metric (dancability, cadence, volume...) by genre

find out general user Spotify preferences including artist, songs and genres most commonly used.

Create functions that call the Spotify API's basic functionalities, such as pulling a song / pulling information about a song.

Graphing accuracy, f1score, initital correlations etc...

At the time of this issue we have a Random Forest that can produce an accuracy in the low to mid 70's. Can a neural net do better?

The distinction here is that it should be given a user the song should be from their library / playlists

Medium article, website, notebook

For facial recognition. We could have the user upload a photo but this would take care of type, location, etc

Pull the Genre for each song