SVHN WideResNet 16-4

class deepobs.svhn.svhn_wrn164.set_up[source]

Function providing the functionality for the WideResNet 16-4 architecture on SVHN.

This function is a wrapper for the svhn_wrn.set_up to create the 16-4 variant of the WideResNet.

The training setting in the paper were: Batch size of 128, weight decay of 0.0005, total training time of 160 epochs, with a learning rate schedule of 0.01, 0.001 after 80 epochs, 0.0001 after 120 epochs. Training was done using Nesterov momentum with a momentum parameter of 0.9.

Parameters:
  • batch_size (int) -- Batch size of the data points.
  • weight_decay (float) -- Weight decay factor. Defaults to 0.0005.
Returns:

Setup class for WideResNets on SVHN, svhn_wrn.set_up.
Return type:

svhn_wrn.set_up