def multilayer_perceptron():
mlnn = MultiLayerNeuralNetwork( [2, 4, 1],
threshold=0.5,
learning_coefficient=0.5,
sigmoid_alpha=5)
x_range = [0,10]
y_range = [0,10]
# liner_data = TrainingData.liner_training_data(x_range, y_range)
#liner_data = TrainingData.quadratic_function_data(x_range, y_range)
liner_data = TrainingData.sin_function_data(x_range, y_range, 5)
train_data = TrainingData.change_format(liner_data)
# 教師データのプロット
fig = plt.figure()
scat(fig, [key for key, value in liner_data.items() if value == 0], color='g' )
scat(fig, [key for key, value in liner_data.items() if value == 1], color='b' )
# 学習
sample_border = len(train_data)
random.shuffle(train_data)
#mlnn.train(train_data[:20])
mlnn.train(train_data[:sample_border])
# xに対応するyを算出
data = get_predict_list(x_range,y_range, mlnn)
# 学習後分離線
plot(fig, data)
plt.show()
def test_multilayer_perceptron_train(self):
# mlnn = MultiLayerNeuralNetwork([2, 2, 1], threshold=0.5, learning_coefficient=0.5, sigmoid_alpha=5)
mlnn = MultiLayerNeuralNetwork([2, 4, 1], threshold=0.5, learning_coefficient=0.5, sigmoid_alpha=5, print_error=False)
#mlnn = MultiLayerNeuralNetwork([2, 4, 4, 1], threshold=0.5, learning_coefficient=0.5, sigmoid_alpha=5)
train_data = [
[[0,0],[1]],
[[1,0],[0]],
[[0,1],[0]],
[[1,1],[1]],
]
# train
mlnn.train(train_data)
# print train_data
# print mlnn.weights
# print mlnn.predict([0, 0])
# print mlnn.predict([1, 0])
# print mlnn.predict([0, 1])
# print mlnn.predict([1, 1])
assert mlnn.predict([0, 0]) > mlnn.predict([1, 0])
assert mlnn.predict([0, 0]) > mlnn.predict([0, 1])
assert mlnn.predict([1, 1]) > mlnn.predict([1, 0])
assert mlnn.predict([1, 1]) > mlnn.predict([0, 1])
def test_multilayer_perceptron_predict(self):
mlnn = MultiLayerNeuralNetwork([2, 1], threshold=1.0, learning_coefficient=1.0, sigmoid_alpha=1.0, print_error=False)
mlnn.weights = {1: np.array([[-1, 1, 1]])}
assert mlnn.predict([0, 0]) == [1 / (1 + math.e ** 1)]
assert mlnn.predict([0, 1]) == [0.5]
assert mlnn.predict([1, 0]) == [0.5]
assert mlnn.predict([1, 1]) == [1 / (1 + math.e ** -1)]
assert mlnn.predict_all_layer([0, 0]) == [[0, 0],[1 / (1 + math.e ** 1)]]
assert mlnn.predict_all_layer([0, 1]) == [[0, 1],[0.5]]
assert mlnn.predict_all_layer([1, 0]) == [[1, 0],[0.5]]
assert mlnn.predict_all_layer([1, 1]) == [[1, 1],[1 / (1 + math.e ** -1)]]
# assert mlnn.predict([0, 0], 1) == [1 / (1 + math.e ** 1)]
# assert mlnn.predict([0, 1], 1) == [0.5]
# assert mlnn.predict([1, 0], 1) == [0.5]
# assert mlnn.predict([1, 1], 1) == [1 / (1 + math.e ** -1)]
# assert mlnn.predict([0, 0], 0) == [0, 0]
# assert mlnn.predict([0, 1], 0) == [0, 1]
# assert mlnn.predict([1, 0], 0) == [1, 0]
# assert mlnn.predict([1, 1], 0) == [1, 1]
# XORが計算できることを重み指定して確認したい.
mlnn.weights = {1: np.array([[-0.5, 1, -1], [-0.5, -1, 1]]) ,
2: np.array([[-0.5, 1, 1],])}