This is implementation of Hybrid Dropout for Pytorch with some benchmarking.
- hybrid_dropout.py - contains implementations of normal and spatial hybrid dropouts
- train_model.py - trains a model using one of the dropout types and saves metric progression to /data/
- Project_exploration.ipynb - a notebook containing some sample results. Also contains initial attempts to use Hybrid Dropout to estimate model uncertainty.
- requirements.txt - if you do pip install -r requirements.txt it should install the Python packages required to run code in this project.