class NetworkTests(unittest.TestCase):
def setUp(self):
self.red = NeuralNetwork(2, [1, 1])
self.layer = NeuralLayer(2, 1)
self.layer2 = NeuralLayer(1, 1)
self.neuron = Neuron(2)
self.neuron2 = Neuron(1)
self.red2 = NeuralNetwork(2, [2, 2])
self.lay1 = NeuralLayer(2, 2)
self.lay2 = NeuralLayer(2, 2)
self.neur1 = Neuron(2)
self.neur2 = Neuron(2)
self.neur3 = Neuron(2)
self.neur4 = Neuron(2)
# testcase entregado en ucursos
def testCase1(self):
self.neuron.custom([0.4, 0.3], 0.5)
self.neuron2.custom([0.3], 0.4)
self.layer.custom([self.neuron])
self.layer2.custom([self.neuron2])
self.red.custom([self.layer, self.layer2])
self.red.train([1, 1], [1])
self.assertEqual(round(self.neuron.weight[0], 15), 0.402101508999489)
self.assertEqual(round(self.neuron.weight[1], 15), 0.302101508999489)
self.assertEqual(round(self.neuron2.weight[0], 15), 0.330262548639919)
self.assertEqual(round(self.neuron.bias, 15), 0.502101508999489)
self.assertEqual(round(self.neuron2.bias, 15), 0.439377453127974)
def testCase2(self):
self.neur1.custom([0.7, 0.3], 0.5)
self.neur2.custom([0.3, 0.7], 0.4)
self.neur3.custom([0.2, 0.3], 0.3)
self.neur4.custom([0.4, 0.2], 0.6)
self.lay1.custom([self.neur1, self.neur2])
self.lay2.custom([self.neur3, self.neur4])
self.red2.custom([self.lay1, self.lay2])
self.red2.train([1, 1], [1, 1])
self.assertEqual(round(self.neur1.weight[0], 15), 0.702510448549328)
self.assertEqual(round(self.neur1.weight[1], 15), 0.302510448549328)
self.assertEqual(round(self.neur2.weight[0], 15), 0.302498011357483)
self.assertEqual(round(self.neur2.weight[1], 15), 0.702498011357483)
self.assertEqual(round(self.neur3.weight[0], 15), 0.229947378819557)
self.assertEqual(round(self.neur3.weight[1], 15), 0.329383628639501)
self.assertEqual(round(self.neur4.weight[0], 15), 0.419430056526462)
self.assertEqual(round(self.neur4.weight[1], 15), 0.219064291698386)
self.assertEqual(round(self.neur1.bias, 15), 0.502510448549328)
self.assertEqual(round(self.neur2.bias, 15), 0.402498011357483)
self.assertEqual(round(self.neur3.bias, 15), 0.336629542251590)
self.assertEqual(round(self.neur4.bias, 15), 0.623765488150905)
# number = random.randint(0, 17)
if number == 1 or number == 2 or number == 3:
rightAnswers = np.array([[1, 0]])
elif number == 4 or number == 5 or number == 6:
rightAnswers = np.array([[0, 1]])
elif 6 < number < 16 or number == 17:
rightAnswers = np.array([[1, 1]])
elif number == 0 or number == 16:
rightAnswers = np.array([[0, 0]])
# print('rightAnswers = ', rightAnswers)
image = cv2.imread("dataset/test/" + str(number) + ".png",
0).flatten() / 255
if pro == 15 and i % 25 == 0:
ppi = np.vstack((ppi, n.train(image, rightAnswers).T))
plt.clf()
plt.xlabel(i * 17)
plt.plot(ppi)
plt.draw(
) # Должно быть это, а в итоге не это, приходится каждый раз открывать картинку
plt.pause(0.0000000000000000000001)
else:
n.train(image, rightAnswers)
plt.title('lr = ' + str(learning_rate))
plt.ioff()
plt.show()
for i in range(18):
image = cv2.imread("dataset/test/" + str(i) + ".png", 0).flatten() / 255
def main():
# Creating NN
nn = NeuralNetwork([784, 64, 3])
# Creating training data
createTraining(cats_training, "cat")
createTraining(trains_training, "train")
createTraining(rainbows_training, "rainbow")
# Creating test data
createTest(cats_testing, "cat")
createTest(trains_testing, "train")
createTest(rainbows_testing, "rainbow")
# make one array from all the training and test data
training_data = cats_training + trains_training + rainbows_training
testing_data = cats_testing + trains_testing + rainbows_testing
# Normalize the training data and testing data, (0-1)
normalizeValues(trains_training)
normalizeValues(testing_data)
# arrange the data in specific format that will fit to the NN library i created.
for j in range(2400):
for i in range(785):
training_data[j][i] = [training_data[j][i]]
for j in range(600):
for i in range(785):
testing_data[j][i] = [testing_data[j][i]]
'''
training the model with the training data (30 times),
shuffle the training data every iteration.
specific target to every label.
'''
for i in range(35):
random.shuffle(training_data)
for j in range(2400):
target = [[0], [0], [0]]
if (training_data[j][784] == ['cat']):
target[0] = [1]
nn.train(training_data[j][:-1], target, 1)
if (training_data[j][784] == ['train']):
target[1] = [1]
nn.train(training_data[j][:-1], target, 1)
if (training_data[j][784] == ['rainbow']):
target[2] = [1]
nn.train(training_data[j][:-1], target, 1)
'''
Calculate the acc. should be more various of training data.
currently : 800 cats, 800 rainbows, 800 trains.
'''
acc = 0
trains = 0
rainbows = 0
cats = 0
for i in range(len(testing_data)):
label = testing_data[i][784]
guess = nn.feed_forword(testing_data[i][:-1])
if max(guess) == guess[0] and label == ['cat']:
acc += 1
cats += 1
if max(guess) == guess[1] and label == ['train']:
acc += 1
trains += 1
if max(guess) == guess[2] and label == ['rainbow']:
acc += 1
rainbows += 1
print("Acc : " + str(acc))
print("length of testing data : " + str(len(testing_data)))
print("Cats : " + str(cats))
print("Rainbows : " + str(rainbows))
print("trains : " + str(trains))
