yimothysu / progressive-rock-classifier

CAP6610 Project: Progressive Rock or Not Progressive Rock?

Progressive Rock Classifier

Install

First, create a virtual environment and install dependencies.

python3.11 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Then, create a directory named data at the project root. Place the provided Not_Progressive_Rock, Progressive_Rock_Songs, and CAP6610SP24_test_set directories in the data directory.

Preprocess

The data is not directly usable yet for model training. Run the following preprocessing scripts to preprocess it into a usable format.

python preprocess.py
python preprocess_test_set.py
python feature_extraction.py

Train

Run python train.py to train the model.

There are two model architectures: CNN and RESNET_18 that you can choose from. CNN is a single convolutional layer (with BatchNorm, Max Pooling, etc.) while RESNET_18 is an 18-layer ResNet as introduced in Kaiming He's 2015 seminal paper. The default model selection is RESNET_18.

Confusion Matrices

Run python snippet_confusion_matrix.py to produce the snippet accuracy and confusion matrix.

Run python song_confusion_matrix.py to produce the song accuracy and confusion matrix.

Other Utilities

Run classify_to_csv.py to generate tables displaying the classification and prediction confidence of each song.

Run compute_statistics.py to get the count and total runtime length of songs in a directory.