def numerical_gradient(self, x, t):
# 勾配
loss_W = lambda W: self.loss(x, t)
grads = {}
grads['W1'] = numerical_gradient(loss_W, self.params['W1'])
grads['b1'] = numerical_gradient(loss_W, self.params['b1'])
grads['W2'] = numerical_gradient(loss_W, self.params['W2'])
grads['b2'] = numerical_gradient(loss_W, self.params['b2'])
return grads
def numerical_gradient(self, x, t):
loss_w = lambda w: self.loss(x, t)
grads = {}
grads['W1'] = functions.numerical_gradient(loss_w, self.params['W1'])
grads['b1'] = functions.numerical_gradient(loss_w, self.params['b1'])
grads['W2'] = functions.numerical_gradient(loss_w, self.params['W2'])
grads['b2'] = functions.numerical_gradient(loss_w, self.params['b2'])
return grads
def numerical_gradient(self, x, t):
def loss_W(W):
return self.loss(x, t)
grads = {}
grads['W1'] = f.numerical_gradient(loss_W, self.params['W1'])
grads['b1'] = f.numerical_gradient(loss_W, self.params['b1'])
grads['W2'] = f.numerical_gradient(loss_W, self.params['W2'])
grads['b2'] = f.numerical_gradient(loss_W, self.params['b2'])
return grads
def numerical_gradient(self, x, t):
loss_w = lambda w: self.loss(x, t)
grads = {}
for idx in (1, 2, 3):
grads['W' + str(idx)] = numerical_gradient(
loss_w, self.params['W' + str(idx)])
grads['b' + str(idx)] = numerical_gradient(
loss_w, self.params['b' + str(idx)])
return grads
def numerical_gradient(self, x, t):
loss_W = lambda W: self.loss(x, t)
grads = {}
for idx in range(1, self.hidden_layer_num + 2):
grads['W' + str(idx)] = numerical_gradient(
loss_W, self.params['W' + str(idx)])
grads['b' + str(idx)] = numerical_gradient(
loss_W, self.params['b' + str(idx)])
return grads
def numerical_gradient(self, x: np.core.multiarray, t: np.core.multiarray):
loss_W = lambda W: self.loss(x, t)
grads = {
'W1': numerical_gradient(loss_W, self.params['W1']),
'b1': numerical_gradient(loss_W, self.params['b1']),
'W2': numerical_gradient(loss_W, self.params['W2']),
'b2': numerical_gradient(loss_W, self.params['b1'])
}
return grads
def numerical_gradient(self, x: np.ndarray, t: np.ndarray) -> np.ndarray:
loss = lambda W: self.calc_loss(x, t)
grads = {}
for key in self.weight.keys():
grads[key] = numerical_gradient(loss, self.weight[key])
return grads
def numerical_gradient(self, x, t):
loss_W = lambda W: self.loss(x, t, train_flg=True)
grads = {}
for idx in range(1, self.hidden_layer_num + 2):
grads['W' + str(idx)] = numerical_gradient(
loss_W, self.params['W' + str(idx)])
grads['b' + str(idx)] = numerical_gradient(
loss_W, self.params['b' + str(idx)])
if self.use_batchnorm and idx != self.hidden_layer_num + 1:
grads['gamma' + str(idx)] = numerical_gradient(
loss_W, self.params['gamma' + str(idx)])
grads['beta' + str(idx)] = numerical_gradient(
loss_W, self.params['beta' + str(idx)])
return grads
def numerical_gradient(self, x, t):
"""Calculate gradient to weight params using numerical gradient.
Args:
x (numpy.ndarray): image data which mean input to NN
t (numpy.ndarray): labels
Return:
dictionary: dictionary of gradient to each param.
"""
def loss_W(W):
return self.loss(x, t)
grads = {}
grads['W1'] = numerical_gradient(loss_W, self.params['W1'])
grads['b1'] = numerical_gradient(loss_W, self.params['b1'])
grads['W2'] = numerical_gradient(loss_W, self.params['W2'])
grads['b2'] = numerical_gradient(loss_W, self.params['b2'])
return grads
def numerical_gradient(self, x, t):
"""勾配を求める（数値微分）
Parameters
----------
x : 入力データ
t : 教師ラベル
Returns
-------
各層の勾配を持ったディクショナリ変数
grads['W1']、grads['W2']、...は各層の重み
grads['b1']、grads['b2']、...は各層のバイアス
"""
loss_w = lambda w: self.loss(x, t)
grads = {}
for idx in (1, 2, 3):
grads['W' + str(idx)] = numerical_gradient(
loss_w, self.params['W' + str(idx)])
grads['b' + str(idx)] = numerical_gradient(
loss_w, self.params['b' + str(idx)])
return grads
def check_numerical_grad():
def func(x):
return x**2
ans = numerical_gradient(func, np.array([[5], [5]]))
print(ans)
from functions import softmax, cross_entropy_error, numerical_gradient
class simpleNet:
def __init__(self):
self.W = np.random.randn(2, 3)
def predict(self, x):
return np.dot(x, self.W)
def loss(self, x, t):
z = self.predict(x)
y = softmax(z)
loss = cross_entropy_error(y, t)
return loss
if __name__ == '__main__':
net = simpleNet()
print(net.W)
x = np.array([0.6, 0.9])
p = net.predict(x)
print(p)
print(np.argmax(p))
t = np.array([0, 0, 1])
def f(W):
return net.loss(x, t)
print(numerical_gradient(f, net.W))
import numpy as np
from functions import numerical_gradient
def function_2(x):
return x[0]**2 + x[1]**2
print(numerical_gradient(function_2, np.array([3.0, 4.0])))
print(numerical_gradient(function_2, np.array([0.0, 2.0])))
print(numerical_gradient(function_2, np.array([3.0, 0.0])))
def predict(self, x):
return np.dot(x, self.W)
def loss(self, x, t):
z = self.predict(x)
y = softmax(z)
loss = cross_entropy_error(y, t)
return loss
def f(W):
return net.loss(x, t)
print("net.W")
net = SimpleNet()
print(net.W)
print("net.predict(x)")
x = np.array([0.6, 0.9])
p = net.predict(x)
print(p)
np.argmax(p)
t = np.array([0, 0, 1])
net.loss(x, t)
print("numerical_gradient")
dW = numerical_gradient(f, net.W)
print(dW)