return (np.sum((t-y)**2))/(len(x))
def error_val(x, t):
y = np.dot(x, W) + b
return (np.sum((t - y) ** 2)) / (len(x))
f = lambda x : loss_func(x_data, t_data)
learning_rate = 1e-3
print("Initial error value =", error_val(x_data, t_data), "Initial W =", W, "\n", ", b =", b)
for step in range(10001):
W -= learning_rate * nd.numerical_derivative(f,W)
b -= learning_rate * nd.numerical_derivative(f,b)
if(step % 400 ==0):
print("step =", step, "error value =", error_val(x_data,t_data), "W =", W, ", b =",b)
e = np.zeros_like(t_data)
predict_velocity = []
def predict():
cnt = 0
delta = 1e-7
y = np.dot(x_data, W) + b
for i in range(len(t_data)):
error_rate = (t_data[i] - y[i]) / (t_data[i] + delta) * 100
print(i,"예측속도: ",y[i],"실제속도: ",t_data[i],"오차율[%]", error_rate
,"W =",W, "b =", b )
predict_velocity.append(y[i])
self.b = np.random.rand(output_nodes)
def loss_func(self):
y = np.dot(self.xdata, self.W) + self.b
return np.sum((self.tdata - y)**2) / len(self.xdata)
def predict(self, xdata):
y = np.dot(xdata, self.W) + self.b
return y
loaded_data = np.loadtxt('./regression_testdata_05.csv',
delimiter=',',
dtype=np.float32)
xdata = loaded_data[:, 0:-1].reshape(-1, 4)
tdata = loaded_data[:, -1].reshape(-1, 1)
learning_rate = 1e-3
obj1 = myLinear(xdata, tdata, 4, 1)
f = lambda x: obj1.loss_func()
for step in range(1000001):
if (step % 400 == 0):
obj1.W -= learning_rate * nd.numerical_derivative(f, obj1.W)
obj1.b -= learning_rate * nd.numerical_derivative(f, obj1.b)
print("step", step, "error value =", obj1.loss_func(), "W =", obj1.W,
"b =", obj1.b)
x = np.array([3, 5, 6, 2]).reshape(1, 4)
real_val = obj1.predict(x)
print(real_val)
z3 = np.dot(a2, self.W3) + self.b3
y = a3 = self.sigmoid(z3)
if (y > 0.5):
result = 1
if (y < 0.5):
result = 0
return y, result
xdata = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])
xor_tdata = np.array([0, 1, 1, 0]).reshape(4, 1)
obj1 = MyDeep(xdata, xor_tdata, 2, 6, 1)
learning_rate = 1e-4
f = lambda x: obj1.feed_forward()
for step in range(10001):
obj1.W2 -= learning_rate * nd.numerical_derivative(f, obj1.W2)
obj1.b2 -= learning_rate * nd.numerical_derivative(f, obj1.b2)
obj1.W3 -= learning_rate * nd.numerical_derivative(f, obj1.W3)
obj1.b3 -= learning_rate * nd.numerical_derivative(f, obj1.b3)
if (step % 400 == 0):
print("step =", step, "loss val =", obj1.feed_forward(), "obj1.W2",
obj1.W2, "obj1.b2", obj1.b2, "obj1.W3", obj1.W3, "obj1.b3",
obj1.b3)
for data in xdata:
(real_val, logical_val) = obj1.predict(data)
print("real_val", real_val, "logical_val", logical_val)
return -np.sum(tdata * np.log(y + delta) + (1 - tdata) * np.log((1 - y) + delta))
def predict(xdata):
z2 = np.dot(xdata, W2) + b2
a2 = sigmoid(z2)
z3 = np.dot(a2,W3) + b3
y = a3 = sigmoid(z3)
if y > 0.5:
result = 1
else:
result = 0
return y, result
f = lambda x : feed_forward(xdata, xor_tdata)
print("Initial loss value = ", loss_val(xdata, xor_tdata))
for step in range(8001):
W2 -= learning_rate + nd.numerical_derivative(f, W2)
b2 -= learning_rate + nd.numerical_derivative(f, b2)
W3 -= learning_rate + nd.numerical_derivative(f, W3)
b3 -= learning_rate + nd.numerical_derivative(f, b3)
if (step % 400 == 0):
print("step =",step,", loss value =", loss_val(xdata, xor_tdata))
for data in test_data:
(real_val, logical_val) = predict(data)
print("real_val", real_val, "logical_val", logical_val)