Example with pytorch-cifar

Note

If Oríon not installed: pip install orion

If the database is not setup, you can follow the instructions here: Setup Database.

Alternatively, you can test the example without setting up a database by using the option –debug, but note that all data gathered during an execution will be lost at the end of it.

pip3 install torch torchvision

git clone git@github.com:kuangliu/pytorch-cifar.git

cd pytorch-cifar

sed -i '1 i\#!/usr/bin/env python' main.py

Add to last line of test()

return 1 - (correct / len(test_loader.dataset))

Last line of the main() function

    test_error_rate = test(epoch)

report_results([dict(
    name='test_error_rate',
    type='objective',
    value=test_error_rate)])

orion -v hunt -n resnet18-cifar10 python main.py --lr~'loguniform(1e-5, 1.0)'

Note

If you are using python3, the script will fail with message

ImportError: No module named 'vgg'

You can fix this with the following command from the repository’s root.

sed -i 's/from /from ./g' models/__init__.py