The repository consists of notebooks containing different experiments, ensemble.py file containing script for ensembling, ensemble_submissions with all the submission files used for final ensembling (model_accuracy > 0.92) and data containing train and test images.
How to setup train folder?
Download the data provided in the competition page. The train.zip consists of train and test images inside train folder. We are also provided with train.csv and test.csv, which will be used to segregate the data between train and test. Place the unzipped train folder inside train/(containing train images and train.csv) and use the test.csv to copy images from train/train/ folder totrain/test/. This will ensure that the data directory is setup to be used with the fastai codebase.
Notebooks
The notebooks inside the notebooks/ directory will act as codebase for various experiments that were conducted.
Ensembling
To generate final ensembling predictions, run:
python ensemble.py