def main():
# training sets for the XOR problem only have 4 cases
training_sets = [
[[0, 0], [0]],
[[0, 1], [1]],
[[1, 0], [1]],
[[1, 1], [0]]
]
# call global variable to use sigmoid for this problem
global activation
activation = "sigmoid"
nn = NeuralNetwork(len(training_sets[0][0]), 5, len(training_sets[0][1]), False)
# can use large learning rate due to simple training
nn.LEARNING_RATE = 1
temp1, temp2 = [], []
# bigger range = longer training = less error = better results
for i in range(10000):
training_inputs, training_outputs = random.choice(training_sets)
nn.train(training_inputs, training_outputs)
temp1.append(i)
k = nn.calculate_total_error(training_sets)
temp2.append(k)
print(i, k)
print()
test1 = [0, 0]
test2 = [0, 1]
test3 = [1, 0]
test4 = [1, 1]
# verify performance
print(nn.feed_forward(test1)[0])
print(nn.feed_forward(test2)[0])
print(nn.feed_forward(test3)[0])
print(nn.feed_forward(test4)[0])
print()
plt.plot(temp1, temp2)
nn.inspect()
training_sets = []
# much larger training set to account for many multiplications
for i in range(1000):
x = random.randint(-100, 100)
y = random.randint(-100, 100)
training_sets.append([[x / 100.0, y / 100.0], [(x * y) / 10000.0]])
# call global variable to use relu for this problem
global activation
activation = "relu"
nn = NeuralNetwork(len(training_sets[0][0]), 20, len(training_sets[0][1]), True)
# smaller learning rate so error rate doesn't stagnate
nn.LEARNING_RATE = .05
temp1, temp2 = [], []
# my computer's not fast enough to run a higher value so my error rate is high
for i in range(10000):
training_inputs, training_outputs = random.choice(training_sets)
nn.train(training_inputs, training_outputs)
temp1.append(i)
k = nn.calculate_total_error(training_sets)
temp2.append(k)
print(i, k)
plot_learning_curves(temp1, temp2)
nn.inspect()
