Guang Yang (gy552) & Chen Zhang (cz1389)
We implemented several models. For your convinence, we placed each banch in a seperate directory.
The directories are name as "GalaxyZoo-"
In each directory,
-- download.sh # A shell script allowing you to download the data used for the task.
# Your need to export your cookie for the Kaggle challenge and put it in cookie.txt.
# Or you can just download the data from the browser.
-- main.py # The skeleton of the training and validation.
-- data.py # Data preparation including preprocessing and augmentation.
-- model.py # The model definition.
-- evaluate.py # Generate the test output.
The files are brought from assignment 3, therefore have similar command line argument.
To train the model, invoke
python main.py --epochs 200 --model_name <model_name>
It will find the local minimums of the validation loss, and dump the model as _.pth
To generate the test output, invoke
python evaluate.py --model _.pth