def main():
numpy.random.seed(3)
if not os.path.exists("cache"):
os.makedirs("cache")
if os.path.exists("cache/training_data.npy"):
training_data = numpy.load('cache/training_data.npy')
else:
training_data = read_images(TRAIN['data'], DATA_FOLDER)
numpy.save('cache/training_data', training_data)
if os.path.exists("cache/training_labels.npy"):
training_labels = numpy.load('cache/training_labels.npy')
else:
training_labels = read_labels(TRAIN['labels'], DATA_FOLDER)
numpy.save('cache/training_labels', training_labels)
training_data = training_data.reshape(training_data.shape[0], -1)
number_of_inputs = training_data.shape[1]
number_of_labels = numpy.unique(training_labels).size
topology = [number_of_inputs, 50, 30, number_of_labels]
if os.path.exists("cache/test_data.npy"):
test_data = numpy.load('cache/test_data.npy')
else:
test_data = read_images(TEST['data'], DATA_FOLDER)
numpy.save('cache/test_data', test_data)
if os.path.exists("cache/test_labels.npy"):
test_labels = numpy.load('cache/test_labels.npy')
else:
test_labels = read_labels(TEST['labels'], DATA_FOLDER)
numpy.save('cache/test_labels', test_labels)
test_data = test_data.reshape(test_data.shape[0], -1)
normalized = Normalizer(training_data=training_data, test_data=test_data)
for i in range(1):
print("Running {}th time...".format(i))
neural_network = NeuralNetwork(topology)
neural_network.train(normalized.training_data(),
training_labels,
test_data=normalized.test_data(),
test_labels=test_labels)