alexandrelam / spotifyml

Neural Network Classifier guesses Spotify songs genre 🧠🎸🤘🎼

Neural Network for Spotify genre
(97% accuracy)

I built a neural network that guesses the genre of a music based on data
scrapped on Spotify!

Report Bug · Request Feature

Built With

• Tensorflow
• Keras
• Scikit-learn
• Pandas
• Numpy

(back to top)

Results

prediction: punk | real result: punk
prediction: country | real result: country

good/total: 28/30 - 93%

(back to top)

Getting Started

Train the neural network

1. Install dependancies

pip install -r requirements.txt

2. Run

python train.py

3. Open the dashboard

tensorboard --logdir=logs

Run the test

1. Install dependancies

pip install -r requirements.txt

2. Change NUMBER_VALUES in test.py

3. Run!

python test.py

(back to top)

Media

prediction: pop | real result: pop
prediction: rock | real result: rock
prediction: pop | real result: pop
prediction: edm | real result: edm
prediction: jazz | real result: jazz
prediction: electro | real result: electro
prediction: pop | real result: pop
prediction: rock | real result: rock
prediction: salsa | real result: salsa
prediction: chanson | real result: chanson
prediction: chanson | real result: hip hop
prediction: rock | real result: rock
prediction: hip hop | real result: hip hop
prediction: dance | real result: dance
prediction: pop | real result: pop
prediction: hip hop | real result: soul
prediction: jazz | real result: jazz
prediction: pop | real result: pop
prediction: hip hop | real result: hip hop
prediction: hip hop | real result: hip hop
prediction: pop | real result: pop
prediction: edm | real result: edm
prediction: dance | real result: dance
prediction: rock | real result: rock
prediction: disco | real result: disco
prediction: rock | real result: rock
prediction: country | real result: country
prediction: blues | real result: blues
prediction: punk | real result: punk
prediction: country | real result: country

good/total: 28/30 - 93%

(back to top)

Roadmap

• with a long training acc should be ok, but test dataset has low acc. maybe it's overtraining...

Things to try

• change the model (layers, categorical to binary?)
• drop some col on the dataset that are useless
• check normalization

(back to top)