Generator that runs a slow source generator in a separate thread. buffer_size: the maximal number of items to pre-generate (length of the buf

Generator that runs a slow source generator in a separate process. buffer_size: the maximal number of items to pre-generate (length of the bu

based on matlab code by Guanglei Xiong, see http://www.mathworks.com/matlabcentral/fileexchange/8303-local-normalization assuming chunk.shape

specify images_gen(cycle(list(train_ids))) as the ids_gen to loop through the training set indefinitely in random order. The shape parameter

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

this generates the training data in order, for postprocessing. Do not use this for actual training.

dump parameters from layer l and down into a file. The dump file has the same name as the script, with _paramdump.pkl appended. This dum

Recursive function to gather weight rescaling updates when the constant is the same for all layers.

in the paper, no learning rate is considered (so learning_rate=1.0). Probably best to keep it at this value. epsilon is important for the ver

epsilon is not included in the typical formula, See "Notes on AdaGrad" by Chris Dyer for more info.

Nesterov momentum, but excluding the biases from the weight decay.

epsilon is not included in Hinton's video, but to prevent problems with relus repeatedly having 0 gradients, it is included here. Watch this

based on matlab code by Guanglei Xiong, see http://www.mathworks.com/matlabcentral/fileexchange/8303-local-normalization

return theano variable

return theano variable