Generate your own valid Kaggle Token and place it under src/data/kaggle.json.

To use the Kaggle API, sign up for a Kaggle account at https://www.kaggle.com. Then go to the 'Account' tab of your user profile (https://www.kaggle.com/<username>/account) and select 'Create API Token'. This will trigger the download of kaggle.json, a file containing your API credentials.

The kaggle.json file is also required to have limited access rights, which can be set through the following:

chmod 600 src/data/kaggle.json

git clone https://github.com/niralpathak/ECE_228_Project.git
sh ECE_228_Project/src/data/setup.sh

The .sh script does the following:

1. Download, extract, and relocate train & test files from Kaggle
2. Install Python dependencies

In src/model/keras/, each main .py script corresponds to a particular network model .py script. For example, main_custom_0.py will train a network specified in net_custom_0.py.

The classes in net_*.py inherit from the Solver class in solver.py for modularity.

cd ECE_228_Project/src/models/keras
python main_*.py --batch=<batch_size> --epochs=<epoch_size> --training_imgs=../../data/train-jpg/ --training_labels=../../data/train_v2.csv --patience=<patience_size> --testing_imgs=../../data/test-jpg/ --sample_submission=../../data/sample_submission_v2.csv --checkpoint_dir=<checkpoint_dir>

python main_custom_0.py --batch=128 --epochs=10 --training_imgs=../../data/train-jpg/ --training_labels=../../data/train_v2.csv --patience=3 --testing_imgs=../../data/test-jpg/ --sample_submission=../../data/sample_submission_v2.csv --checkpoint_dir=./checkpoint_main