def main():
patterns = loadData('pict.dat') # Pattern 1-11
patterns_1_3 = [patterns[index, :].reshape(1, 1024) for index in range(3)]
patterns_4_11 = [
patterns[3 + index, :].reshape(1, 1024) for index in range(8)
]
network = RNN(size=1024, sequential=False, random=False)
network.init_weights(patterns_1_3)
noises = np.arange(0, 100, 5)
averages = 1000
for i, pattern in enumerate(patterns_1_3):
OGpattern = pattern.copy()
nCorrect = np.zeros((noises.shape[0], 1))
for k, noise in enumerate(noises):
for j in range(averages):
patternD = distort(OGpattern, noise)
x_output = network.train(patternD)
nCorrect[k][0] += ((np.count_nonzero(x_output == OGpattern)) /
patternD.shape[1]) * 100
nCorrect = nCorrect / averages
plt.plot(noises, nCorrect, label=("Pattern " + str(i + 1)))
plt.legend()
plt.show()
def main():
patterns = loadData('pict.dat') # Pattern 1-11
patterns_1_3 = [patterns[index,:].reshape(1,1024) for index in range(3) ]
patterns_4_11 = [patterns[3+index,:].reshape(1,1024) for index in range(8) ]
network = RNN(size=1024, sequential=False, random=True)
network.init_weights(patterns_1_3)
for index, pattern in enumerate(patterns_1_3):
energi = network.layapunovFunction(pattern)
print('Energi for pattern {}: {}'.format(index, energi))
for index, pattern in enumerate(patterns_4_11):
energi = network.layapunovFunction(pattern)
print('Energi for distorted pattern {}: {}'.format(3 + index, energi))
def main():
patterns = loadData('pict.dat') # Pattern 1-11
patterns_1_3 = [patterns[index,:].reshape(1,1024) for index in range(3) ]
patterns_10_11 = [patterns[9+index,:].reshape(1,1024) for index in range(2) ]
network = RNN(size=1024, sequential=True, random=True)
network.init_weights(patterns_1_3)
# Testing if stable
print('\nTesting if stable: ')
plt.figure('Attractors - patterns')
index = 0
for pattern in patterns_1_3:
index += 1
plt.subplot(1, 3, index)
plt.title('Pattern: {}'.format(index))
x_output = network.train(pattern)
print('Number of correct: {}/{} '.format(np.count_nonzero(x_output==pattern), pattern.shape[1]))
plt.imshow(pattern.reshape(32,32), cmap='gray')
# Testing for distorted patterns 9 and 10
print('\nTesting for distorted patterns: ')
index = 10
for pattern in patterns_10_11:
plt.figure('Output - pattern: {}'.format(index))
index += 1
sub_index = 0
x_output = network.train(pattern)
for true_pattern in patterns_1_3:
sub_index += 1
print('Number of correct: {}/{} '.format(np.count_nonzero(x_output==true_pattern), true_pattern.shape[1]))
plt.imshow(x_output.reshape(32,32), cmap='gray')
plt.show()
