truth_count = 0 for p in row: if p : truth_count += 1 return 1 * (truth_count == 1) def truthtable (n): if n < 1: return [[]] subtable = truthtable(n-1) return [ row + [v] for row in subtable for v in [0,1] ] def interpret(v): if v[0] > v[1]: return 1 else: return 0 a = neuralnetwork.makeNNet(3, 3, 3, act.relu, act.relu) b = neuralnetwork.makeNNet(3, 2, 1, act.relu, act.relu) deepnet = neuralnetwork.DeepNet([a, b]) inputs = truthtable(3) targets = [xor(x) for x in inputs] for i in range(3000): for x, t in zip(inputs, targets): deepnet.learn(x, t, 0.01) for x, t in zip(inputs, targets): print t, x, deepnet.feedForward(x)
max_index = 0
for i, n in enumerate(numbers):
if n > numbers[max_index]:
max_index = i
return max_index
path = "/home/lance/git/nnetevolution/src/data"
# Images are 28x28 pixels. Reshapen, they are 784x1
training = [(mux(label), image.reshape(784)/255.0) for label, image in mnist.read(path=path, dataset="training")]
testing = [(mux(label), image.reshape(784)/255.0) for label, image in mnist.read(path=path, dataset="testing")]
a = neuralnetwork.makeSquareConvolution(2, 14)
# pudb.set_trace()
b = neuralnetwork.makeSquareConvolution(2, 7)
nnet = neuralnetwork.makeNNet(49, 10, 10, act.sigmoid, act.relu)
deepnet = neuralnetwork.DeepNet([a, b, nnet])
l = len(training)
k = 1
p = printing.printer(1)
for t, x in training:
deepnet.learn(x, t)
p.reprint("iteration: " + str(k) + "/" + str(l))
k += 1
print
print "finished..."
count = 0
for t, x in testing:
y = demux(deepnet.feedForward(x))
import neuralnetwork from evolution import getAncestor, getDescendant, addNode, splitArc, copy import networkx as nx import matplotlib.pyplot as plt import pylab import pudb def makeTuple(arc): return (str(arc.parent.innovation_number), str(arc.child.innovation_number)) nnet = neuralnetwork.makeNNet(2, 2, 1) for i in range(1): n = addNode(nnet.input_layer + nnet.hidden_layer + nnet.output_layer) nnet.hidden_layer.append(n) # n = splitArc(nnet.input_layer + nnet.hidden_layer + nnet.output_layer) # nnet.hidden_layer.append(n) G = nx.DiGraph() arcs = [] for n in nnet.nodes: for a in n.incoming: arcs.append(makeTuple(a)) G.add_edges_from(arcs) pos = nx.spring_layout(G) nx.draw(G, pos) nx.draw_networkx_labels(G, pos) plt.show()