alrojo / lasagne_residual_network Goto Github PK

@alrojo
this seems quite exciting and interesting to try out. but how to use this? just install Lasagne (dependencies) and run the Python code here?

also which training data to use and where to place that?

will mostly try this out on a AWS GPU running 14.04 Ubuntu.

if you please point out to any key steps it would be really wonderful. thanks!

I just run the demo according to the README.md and got a error:
Amount of bottlenecks: 8
Loading data...
Building model and compiling functions...
Traceback (most recent call last):
File "Deep_Residual_Network_mnist.py", line 493, in
main(**kwargs)
File "Deep_Residual_Network_mnist.py", line 336, in main
network = build_cnn(input_var, n, num_filters)
File "Deep_Residual_Network_mnist.py", line 96, in build_cnn
expression = lasagne.layers.ExpressionLayer
AttributeError: 'module' object has no attribute 'ExpressionLayer'

In original paper, normalization was used before activation, but it seem you have used after activation.
@alrojo Am I wrong?