Judge a Book by its Cover

Kaggle competition: Predict a book's genre given its cover image and title.

Team

• Burouj Armgaan
• Mohit Sharma

Data

Find data.zip at the following link.

Conda environment

• Run conda env create -f env.yaml to create a conda env by the name perk.
• Run conda activate perk prior to running any scripts in this repo.

Expedted dataset_dir structure

• <dataset_dir>
  • images/images/
  • comp_test_x.csv
  • non_comp_test_x.csv
  • non_comp_test_y.csv
  • train_x.csv
  • train_y.csv

Run

• All scripts must be run from the code directory.
• CNN : python cnn.py <dataset_dir_path>
• Bidirectional-RNN : python rnn.py <dataset_dir_path>
• Multimodal : python comp.py <dataset_dir_path>

Description

Running any of the .py files will do the following:

• Train the model
• Print out the test accuracy
• Save the model to disk in the format <model>_<time>.pt.
• Store the test-set predictions in a .csv file named:
  • non_comp_test_pred_y.csv for cnn.py & rnn.py.
  • comp_test_y.csv for comp.py.