Song Genre and Hit Prediction

This repository contains the files of the project Song Genre and Hit Prediction. The structure of this repository with all the files are present in the ee660_project_dirtree document.

Instructions to execute the code

1. Run the download_models.shscript to download the models.
2. Execute the main.py file with optional arguments as shown below to construct the dataset or to train the models.

• The main.py file calls the appropriate file for performing the respective function. For example, if the user input an argument for constructing the dataset, then the dataset_constructor.py file is executed.
• Same is the case for the file train.py which trains the models with the dataset constructed in the previous step

3. The primary function of main.py is to run the trained models downloaded in step 1 on the validation data to perform model selection based on their performance. To do this, the main.py file executes validation.py which evaluates and selects the best model based on the evaluation.
4. The objective is to see how well a model can generalize its predictions, and this is done by test.py file which the main.py file executes to evaluate the models on unseen out of sample data.

Example execution commands

Step 1:

$ bash download_models.sh

Step 2: To construct the dataset:

$ python3 main.py -c True

To train the models on previously constructed datasets:

$ python3 main.py -t True

To construct the dataset and train the models on the newly constructed datasets:

$ python3 main.py -c True -t True

Step 3: To validate the models and choose the best model and evaluate its performance:

$ python3 main.py

The evaluation results will be displayed on the command line directly, and the intermediate and final results with the evaluation metrics will be stored in the results directory.

Additionally the constructed datasets before feature extraction are provided in the data_files.tar.gz compressed file