Install Neon

	git clone https://github.com/NervanaSystems/neon.git
	cd neon
	make
	. .venv/bin/activate
		    

Import required libraries

	import cPickle, numpy as np 
	from neon.initializers import Uniform, Gaussian 
	from neon.layers import GeneralizedCost, Affine 
	from neon.models import Model 
	from neon.optimizers import GradientDescentMomentum, Schedule 
	from neon.transforms import Softmax, CrossEntropyMulti, SumSquared, Rectlin 
	from neon.callbacks.callbacks import Callbacks 
	from neon.data import ArrayIterator 
	from neon.util.argparser import NeonArgparser 
	from neon.transforms import Misclassification 
	from numpy import genfromtxt 
	parser = NeonArgparser(__doc__) 
	args = parser.parse_args(gen_be=True)
			

Create training and test dataset

X_data = genfromtxt("inputXOR.csv",delimiter=',') 
Y_data = genfromtxt("outputXOR.csv",delimiter=',',dtype=None) 

training_data = ArrayIterator(X=X_data[0:600], y=Y_data[0:600],nclass=2,make_onehot=True) 
test_data = ArrayIterator(X=X_data[600:800], y=Y_data[600:800],nclass=2,make_onehot=True)

Build a network by concatenating layers

init = Gaussian() 
layers = [] 
layers.append(Affine(nout=2, init=init, bias=init, activation=Rectlin())) 
layers.append(Affine(nout=2, init=init, bias=init, activation=Softmax()))

Pass layers in to Model class

mlp = Model(layers=layers) 
	

Why Rectilinear Unit?

Why SoftMax?

Select a loss function

cost = GeneralizedCost(costfunc=SumSquared()) 
	

Implement Stochastic Gradient Descent optimizer

optimizer = GradientDescentMomentum(0.1, momentum_coef=0.2) 
callbacks = Callbacks(mlp, eval_set=training_data, **args.callback_args) 

	

Train the Model

mlp.fit(training_data, optimizer=optimizer, num_epochs=100, cost=cost,callbacks=callbacks) 

	

Get predictions and measure accuracy

results = mlp.get_outputs(test_data) 
prediction = results.argmax(1) 
print "Evaluating the model" 
error_pct = 100 * mlp.eval(test_data, metric=Misclassification()) 
print 'Misclassification error = %.1f%%' % error_pct

np.savetxt("predictions.csv", prediction, delimiter=",")

#downloads weights is required
weights =  mlp.get_description(get_weights=True) 
print mlp.get_description(True)['model']['config']['layers'][0]['params']['W'] 
print mlp.get_description(True)['model']['config']['layers'][1]['params']['W']