CS109B Project Group 29
Calvin J Chiew, Ji Hua, Tim Hagmann
https://github.com/greenore/deep-learning-project/blob/master/milestone5/Report.pdf
https://www.youtube.com/watch?v=oyXUTy3y2-k
https://www.dropbox.com/sh/qt20u1fs2jcf9md/AABEYIxwYyTGG4h7mbrX3Evka?dl=0
(Warning: very large file size!)
- Milestone 1: https://github.com/greenore/deep-learning-project/blob/master/milestone1/Milestone1.ipynb
- Milestone 2: https://github.com/greenore/deep-learning-project/blob/master/milestone2/Milestone2.ipynb
- Milestone 3: https://github.com/greenore/deep-learning-project/blob/master/milestone3/Milestone3.ipynb
- Milestone 4: https://github.com/greenore/deep-learning-project/blob/master/milestone4/Milestone4.ipynb
- Milestone 5: https://github.com/greenore/deep-learning-project/blob/master/milestone5/Milestone5.ipynb
- No changes in Milestones 1 and 2
- Milestone 3: Added parameter tuning via cross-validation for random forest and kNN models. Generated new visualisations showing precision and recall of each model by genre.
- Milestone 4: Added section exploring predicted probabilities from the CNN. Attempted to tune the optimal predicted probability thresholds for genre classification instead of using output layer.
- Milestone 5: New notebook showing our attempts to use alternative methods to train the network in a bid to improve its precision and recall. Methods attempted include multi-class classification, binary classification for each genre, and multi-label classification with equal sample sizes per class.