Classifies songs into genres based on feature vectors consisting of 12 features. This is assignment 2 of the ECSE-526 class, as described here.

The program depends on the the pandas and numpy libraries to load and manipulate the song data, as well as the scikit-learn library if using the k-d tree for kNN or any classifiers besides Gaussian and kNN. The matplotlib library was used for plotting. All of these libraries can be installed via pip.

The default data path is song_data/. This is where the test and training directories should be, as well as the labels.csv file. This data can be obtained from Kaggle. Note that the data path can be configured with the command-line arguments described in the next section, although the program will always assume that the test and training directories and labels.csv file are within that path.

Here are some examples of running the program:

Training with a Gaussian classifier:

python main.py --classifier gaussian train

Training with a kNN classifier (k = 1) and using cross-validation with 10 subsets:

python main.py --classifier knn -k 1 train --k_fold 10

Predicting the genres of new songs using the Gaussian classifier:

python main.py --classifier gaussian predict

To run the program, use the command-line interface in main.py. To see the list of available commands, run the following:

python main.py --help

This will print the following:

usage: main.py [-h] [-c {knn,svm,nn,qda,gaussian,gpc,ada}]
               [-l {info,debug,critical,warn,error}] [-d DATA_PATH]
               [-k K_NEAREST] [-s {simple,kd_tree}]
               {train,predict} ...

Music Genre Classification.

positional arguments:
  {train,predict}
    train               Train with training data and test data.
    predict             Predict on new data. Saves the result to CSV in the data path.

optional arguments:
  -h, --help            show this help message and exit
  -c {knn,svm,nn,qda,gaussian,gpc,ada}, --classifier {knn,svm,nn,qda,gaussian,gpc,ada}
                        The classifier to use.
  -l {info,debug,critical,warn,error}, --logging_level {info,debug,critical,warn,error}
                        The logging level.
  -d DATA_PATH, --data_path DATA_PATH
                        The path containing the 'test' and 'training'
                        directories, as well as the 'labels' CSV.
  -k K_NEAREST, --k_nearest K_NEAREST
                        The number of nearest neighbours to use for kNN.
  -s {simple,kd_tree}, --knn_data_structure {simple,kd_tree,average}
                        The data structure to store previous examples for kNN.

To see help relating to the train argument, run the following:

python main.py train --help

This will print the following:

usage: main.py train [-h] [-f K_FOLD]

optional arguments:
  -h, --help            show this help message and exit
  -f K_FOLD, --k_fold K_FOLD
                        The number of partitions to use for k-fold cross-
                        validation.

The predict argument has no further optional arguments.

Here are the default values for all the optional arguments:

The code relating to all the different classifiers used can be found in the classifiers.py file. The main.py file contains the main program, where the command-line arguments are parsed and training or predicting is performed.

The report (report.pdf) and all related files (logs and CSV files) can be found in the report directory.