ds-ga-1008-a2

NYU course 2016 Spring 2016 Assignment 2

This set of code is modified from Sergey Zagoruyko's cifar.torch tutorial: cifar.torch.

Data processing:

th -i provider.lua

provider = Provider()
provider:normalize()
torch.save('provider.t7',provider)

Training:

th train.lua --model vgg_bn_drop -s logs/vgg

Note: cross-validation is not required for this assignment. But please keep in mind that if by any chance you tend to make formal comparison with published results in the future, reading carefully the standard testing protocol is where you should start with.

Acknowledgement

We thank Xiang Zhang for providing a preprocessed STL-10 dataset.