def predict(network, x, mode="INT_MODE"):
W1, W2, W3 = network['W1'], network['W2'], network['W3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
if mode == "FLOAT_MODE":
a1 = np.dot(x, W1)
a1 += b1
z1 = sigmoid(a1)
a2 = np.dot(z1, W2)
a2 += b2
z2 = sigmoid(a2)
a3 = np.dot(z2, W3)
a3 += b3
y = softmax(a3)
else:
a1 = np.dot(x, W1)
a1 = (a1 // P_INT) + b1
a1 = np.clip(a1, -(1 << (P_INT_BITS + 4)), (1 << (P_INT_BITS + 4)) - 1)
z1 = sigmoid_int(a1)
a2 = np.dot(z1, W2)
a2 = (a2 // P_INT) + b2
a2 = np.clip(a2, -(1 << (P_INT_BITS + 4)), (1 << (P_INT_BITS + 4)) - 1)
z2 = sigmoid_int(a2)
a3 = np.dot(z2, W3)
a3 = (a3 // P_INT) + b3
a3 = np.clip(a3, -(1 << (P_INT_BITS + 4)), (1 << (P_INT_BITS + 4)) - 1)
y = softmax(a3)
return y, (a1, a2, a3)
def predict(network, x):
W1, W2, W3 = network['W1'], network['W2'], network['W3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
#pprint_elm('x', x)
#pprint_elm('W1', W1)
#pprint_elm('b1', b1)
a1 = np.dot(x, W1) + b1
#pprint_elm('a1', a1)
z1 = sigmoid(a1)
#pprint_elm('z1', z1)
#pprint_elm('W2', W2)
#pprint_elm('b2', b2)
a2 = np.dot(z1, W2) + b2
#pprint_elm('a2', a2)
z2 = sigmoid(a2)
#pprint_elm('z2', z2)
#pprint_elm('W3', W3)
#pprint_elm('b3', b3)
a3 = np.dot(z2, W3) + b3
#pprint_elm('a3', a3)
y = softmax(a3)
#pprint_elm('y', y)
return y
def predict(network, x):
'''
推論を行う
Parameters
----------
network
x
Returns
-------
'''
W1, W2, W3 = network['W1'], network['W2'], network['W3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
a1 = np.dot(x, W1) + b1
logger.debug(a1)
z1 = sigmoid(a1)
a2 = np.dot(z1, W2) + b2
z2 = sigmoid(a2)
a3 = np.dot(z2, W3) + b3
y = softmax(a3)
return y
def forward(self, x, h_prev, c_prev):
Wx, Wh, b = self.params
N, H = h_prev.shape
# アフィン変換
A = np.dot(x, Wx) + np.dot(h_prev, Wh) + b
# slice
f = A[:, :H]
g = A[:, H:2 * H]
i = A[:, 2 * H:3 * H]
o = A[:, 3 * H:]
f = sigmoid(f)
g = np.tanh(g)
i = sigmoid(i)
o = sigmoid(o)
# 次の時刻の記憶セル:cの値を算出
c_next = f * c_prev + g * i
# 次の時刻の隠れ状態:hの値を算出
h_next = o * np.tanh(c_next)
self.cache = (x, h_prev, c_prev, i, f, g, o, c_next)
return h_next, c_next
def forward(self, x, h_prev):
# x (N, D) = (バッチサイズ, 1時刻あたりの特徴ベクトルの次元)
# h_prev (N, H) = (バッチサイズ, 1時刻あたりの隠れベクトルの次元)
# Wx (D, 3H)
# Wh (H, 3H)
# Wx_z (D, H)
# Wx_r (D, H)
# Wx_h (D, H)
# Wh_z (H, H)
# Wh_r (H, H)
# Wh_h (H, H)
Wx, Wh, b = self.params
H = Wh.shape[0]
Wxz, Wxr, Wxh = Wx[:, :H], Wx[:, H:2*H], Wx[:, 2*H:]
Whz, Whr, Whh = Wh[:, :H], Wh[:, H:2*H], Wh[:, 2*H:]
bz, br, bh = b[:, :H], b[:, H:2*H], b[:, 2*H:]
z = sigmoid(np.dot(x, Wxz) + np.dot(h_prev, Whz) + bz)
r = sigmoid(np.dot(x, Wxr) + np.dot(h_prev, Whr) + br)
h_hat = np.tanh(np.dot(x, Wxh) + np.dot(r*h_prev, Whh) + bh)
h_next = (1 - z) * h_prev + z * h_hat
self.cache = (x, h_prev, r, z, h_hat)
return h_next
def forward(self, x, h_prev, c_prev):
# 순전파
# 현 시각의 입력 x, 이전 시각의 은닉 상태 h_prev, 이전 시각의 기억셀 c_prev
Wx, Wh, b = self.params
N, H = h_prev.shape
A = np.dot(x, Wx) + np.dot(h_prev, Wh) + b
# slice
# 아핀 변환의 결과 행렬을 균등하게 4조각으로 나눠서 꺼내주는 단순한 노드
f = A[:, :H]
g = A[:, H:2*H]
i = A[:, 2*H:3*H]
o = A[:, 3*H:]
# 활성화 함수 (sigmoid 또는 tanh)
f = sigmoid(f)
g = np.tanh(g)
i = sigmoid(i)
o = sigmoid(o)
c_next = f * c_prev + g * i
h_next = o * np.tanh(c_next)
self.cache = (x, h_prev, c_prev, i, f, g, o, c_next)
return h_next, c_next
def forward(self, x, h_prev, c_prev):
Wx, Wh, b = self.params
N, H = h_prev.shape
# 4つ分の重みをまとめてアフィン変換
A = np.dot(x, Wx) + np.dot(h_prev, Wh) + b
# そしてそれぞれを分割
f = A[:, :H]
g = A[:, H:2*H]
i = A[:, 2*H:3*H]
o = A[:, 3*H:4*H]
# ゲート計算
f = sigmoid(f)
g = np.tanh(g)
i = sigmoid(i)
o = sigmoid(o)
# 伝播
c_next = f * c_prev + g * i
h_next = o * np.tanh(c_next)
self.cache = (x, h_prev, c_prev, i, f, g, o, c_next)
return h_next, c_next
def predict(self, x):
W1, W2 = self.params['W1'], self.params['W2']
b1, b2 = self.params['b1'], self.params['b2']
a1 = np.dot(x, W1) + b1
z1 = sigmoid(a1)
a2 = np.dot(z1, W2) + b2
y = sigmoid(a2)
return y
def predict(network, x):
W1, W2, W3 = network['W1'], network['W2'], network['W3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
a1 = np.dot(x, W1) + b1
z1 = sigmoid(a1)
a2 = np.dot(z1, W2) + b2
z2 = sigmoid(a2)
a3 = np.dot(z2, W3) + b3
y = softmax(a3)
return y
def predict(network, x):
W1, W2, W3 = network["W1"], network["W2"], network["W3"]
b1, b2, b3 = network["b1"], network["b2"], network["b3"]
a1 = np.dot(x, W1) + b1
z1 = sigmoid(a1)
a2 = np.dot(z1, W2) + b2
z2 = sigmoid(a2)
a3 = np.dot(z2, W3) + b3
y = softmax(a3)
return y
def forward(network, x):
W1, W2, W3 = network['W1'], network['W2'], network['W3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
a1 = np.dot(x, W1) + b1
z1 = sigmoid(a1)
a2 = np.dot(z1, W2) + b2
z2 = sigmoid(a2)
a3 = np.dot(z2, W3) + b3
y = a3
return y
def forward(network, x):
w1, w2, w3 = network['w1'], network['w2'], network['w3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
a1 = np.dot(x, w1) + b1
z1 = sigmoid(a1)
a2 = np.dot(z1, w2) + b2
z2 = sigmoid(a2)
a3 = np.dot(z2, w3) + b3
y = identity_function(a3)
return y
def predict(network, x):
#Access the parameters and store them in variables.
W1, W2, W3 = network["W1"], network["W2"], network["W3"]
b1, b2, b3 = network["b1"], network["b2"], network["b3"]
a1 = np.dot(x, W1) + b1
z1 = sigmoid(a1)
a2 = np.dot(z1, W2) + b2
z2 = sigmoid(a2)
a3 = np.dot(z2, W3) + b3
y = softmax(a3)
return y
def predict(network, x):
w1, w2, w3 = network['W1'], network['W2'], network['W3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
a1 = np.dot(x, w1) + b1
z1 = sigmoid(a1)
a2 = np.dot(z1, w2) + b2
z2 = sigmoid(a2)
a3 = np.dot(z2, w3) + b3
y = softmax(a3)
return y
def predict_sigmoid(network, x):
w1, w2, w3 = network['W1'], network['W2'], network['W3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
a1 = np.dot(x, w1) + b1
z1 = sigmoid(a1)
a2 = np.dot(z1, w2) + b2
z2 = sigmoid(a2)
a3 = np.dot(z2, w3) + b3
y = softmax(a3)
return y
def forword(network, x):
W1, W2, W3 = network['W1'], network['W2'], network['W3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
a1 = np.dot(x, W1) + b1
z1 = sigmoid(a1)
a2 = np.dot(z1, W2) + b2
z2 = sigmoid(a2)
a3 = np.dot(z2, W3) + b3
y = identity_function(a3)
return y
def forward(self, x, h_prev):
H, H3 = self.Wh.shape
Wxz, Wxr, Wx = self.Wx[:, :H], self.Wx[:, H:2 * H], self.Wx[:, 2 * H:]
Whz, Whr, Wh = self.Wh[:, :H], self.Wh[:, H:2 * H], self.Wh[:, 2 * H:]
z = sigmoid(np.dot(x, Wxz) + np.dot(h_prev, Whz))
r = sigmoid(np.dot(x, Wxr) + np.dot(h_prev, Whr))
h_hat = np.tanh(np.dot(x, Wx) + np.dot(r*h_prev, Wh))
h_next = (1-z) * h_prev + z * h_hat
self.cache = (x, h_prev, z, r, h_hat)
return h_next
def predict(network, x):
'''模型的推理(前向传播)'''
W1, W2, W3 = network['W1'], network['W2'], network['W3'] #取出权重参数
b1, b2, b3 = network['b1'], network['b2'], network['b3'] #取出偏置
a1 = np.dot(x, W1) + b1 #输入（第0层）到第1层的计算
z1 = sigmoid(a1) #使用激活函数非线性化
a2 = np.dot(z1, W2) + b2 #第1层到第2层的计算
z2 = sigmoid(a2)
a3 = np.dot(z2, W3) + b3 #第2层到输出层
y = softmax(a3) #使用softmax处理输出，转换成了概率
return y
def forward(network, x):
'''前向传播'''
W1, W2, W3 = network['W1'], network['W2'], network['W3'] #从字典取出参数
b1, b2, b3 = network['b1'], network['b2'], network['b3']
a1 = np.dot(x, W1) + b1 #利用点积进行从输入层到第1层的信号传递
z1 = sigmoid(a1) #使用激活函数转换第1层的输出
a2 = np.dot(z1, W2) + b2 #同上
z2 = sigmoid(a2) #同上
a3 = np.dot(z2, W3) + b3
y = identity_function(a3) #这里使用恒等函数，只是为了和上面保持格式一致
return y #返回网络的计算结果
def predict(network, x):
W1, W2, W3 = network['W1'], network['W2'], network['W3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
# 由于现在使用了batch，因此输入的x就是100x784，而W1是784x50，所以不用转置
a1 = np.dot(x, W1) + b1
z1 = sigmoid(a1)
a2 = np.dot(z1, W2) + b2
z2 = sigmoid(a2)
a3 = np.dot(z2, W3) + b3
y = softmax(a3)
return np.argmax(y, axis=1)
def predict(network, x): # 예측값을 출력하는 함수
W1, W2, W3 = network['W1'], network['W2'], network[
'W3'] # network에서 가져온 가중치를 각 변수마다 할당
b1, b2, b3 = network['b1'], network['b2'], network[
'b3'] # network에서 가져온 바이어스를 각 변수마다 할당
a1 = np.dot(x, W1) + b1 # x와 w1을 내적한 값에 b1을 더하여 a1에 저장
z1 = sigmoid(a1) # softmax 함수를 통과한 a1을 z1에 저장
a2 = np.dot(z1, W2) + b2 # z1과 w2을 내적한 값에 b2을 더하여 a2에 저장
z2 = sigmoid(a2) # softmax 함수를 통과한 a2를 z2에 저장
a3 = np.dot(z2, W3) + b3 # z2와 w3을 내적한 값에 b3을 더하여 a3에 저장
y = softmax(a3) # softmax 함수를 통과한 a3을 y에 저장
return y # 값을 반환
def predict(network, x):
w1, w2, w3 = network["W1"], network["W2"], network["W3"]
b1, b2, b3 = network["b1"], network["b2"], network["b3"]
a1 = np.dot(x, w1) + b1
z1 = sigmoid(a1)
a2 = np.dot(z1, w2) + b2
z2 = sigmoid(a2)
a3 = np.dot(z2, w3) + b3
z3 = sigmoid(a3)
y = softmax(a3)
return y
def forward(self, x, h_prev):
Wx, Wh, b = self.params
H = Wh.shape[0]
Wxz, Wxr, Wxh = Wx[:, :H], Wx[:, H:2 * H], Wx[:, 2 * H:]
Whz, Whr, Whh = Wh[:, :H], Wh[:, H:2 * H], Wh[:, 2 * H:]
bz, br, bh = b[:H], b[H:2 * H], b[2 * H:]
z = sigmoid(np.dot(x, Wxz) + np.dot(h_prev, Whz) + bz)
r = sigmoid(np.dot(x, Wxr) + np.dot(h_prev, Whr) + br)
h_hat = np.tanh(np.dot(x, Wxh) + np.dot(r*h_prev, Whh) + bh)
h_next = (1-z) * h_prev + z * h_hat
self.cache = (x, h_prev, z, r, h_hat)
return h_next
def predict(self, x):
# 1層目
W1 = self.params['W1']
b1 = self.params['b1']
a1 = np.dot(x, W1) + b1
z1 = sigmoid(a1)
# 2層目
W2 = self.params['W2']
b2 = self.params['b2']
a2 = np.dot(z1, W2) + b2
z2 = sigmoid(a2)
# 出力層
return softmax(z2)
def forward(network, x):
print ("##### 順伝播開始 #####")
W1, W2, W3 = network ['W1'], network ['W2'], network ['W3']
b1, b2, b3 = network ['b1'], network ['b2'], network ['b3']
# 隠れ層の総入力
u1 = np.dot (x, W1) + b1
# 隠れ層1の総出力
z1 = functions.relu (u1)
# 隠れ層２層への総入力
u2 = np.dot (z1, W2) + b2
# 隠れ層2の出力
z2 = functions.relu (u2)
u3 = np.dot (z2, W3) + b3
z3 = functions.sigmoid (u3)
y = z3
print_vec ("総入力1", u1)
print_vec ("中間層出力1", z1)
print_vec ("総入力2", u2)
print_vec ("出力1", y)
print ("出力合計: " + str (np.sum (y)))
return y, z1
def predict(network, x):
W1, W2, W3 = network['W1'], network['W2'], network['W3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
# a1 = np.dot(x, W1) + b1
# z1 = sigmoid(a1)
# a2 = np.dot(z1, W2) + b2
# z2 = sigmoid(a2)
# a3 = np.dot(z2, W3) + b3
# y = softmax(a3)
z1 = sigmoid(np.dot(x, W1) + b1)
z2 = sigmoid(np.dot(z1, W2) + b2)
y = softmax(np.dot(z2, W3) + b3)
return y
def forward(self, x, t):
self.t = t
self.y = sigmoid(x)
y = np.c_[1.0 - self.y, self.y]
loss = cross_entropy_error(y, self.t)
return loss
def predict(network, x):
B1, B2, B3 = network['b1'], network['b2'], network['b3']
W1, W2, W3 = network['W1'], network['W2'], network['W3']
a1 = np.dot(x, W1) + B1
z1 = sigmoid(a1)
# print(z1.shape)
a2 = np.dot(z1, W2) + B2
z2 = sigmoid(a2)
a3 = np.dot(z2, W3) + B3
y = softmax(a3)
return y
def CompletePlot(self,
time=0,
save=False,
start_time=9,
path='~/Data/ShadeRatio/Machida3h/tmp'):
xyt = np.empty((0, self.dimension + 1), float)
for x in self.xgrid:
for y in self.ygrid:
xyt = np.append(xyt, np.array([[x, y, time]]), axis=0)
self.LayerBuild()
q = np.array([self.predict(x) \
for x in xyt])
z = sigmoid(self.a * q + self.b)
zgrid = z.reshape(self.grid[0], self.grid[1])
sns.heatmap(zgrid,
annot=False,
vmin=0,
vmax=1,
fmt='g',
cmap='YlGnBu_r')
if (save == True):
sns.plt.savefig(path + str(start_time) + str(2.5 * time) + '.png')
sns.plt.show()
def Plot(self, frame=np.arange(5), axtype='contourf'):
self.LayerBuild()
q = np.array([self.predict(x) \
for x in self.xyt])
frame_mask = np.array([self.xyt[:, 2] == f \
for f in self.frame])
x = self.xyt[frame_mask[0], 0]
y = self.xyt[frame_mask[0], 1]
Z = sigmoid(self.a * q + self.b)
xgrid = x.reshape(self.grid[0], self.grid[1])
ygrid = y.reshape(self.grid[0], self.grid[1])
sns.set_palette('YlGnBu_r')
for f in frame:
fig = plt.figure()
ax = Axes3D(fig)
z = Z[frame_mask[f]]
zgrid = z.reshape(self.grid[0], self.grid[1])
if (axtype == 'wireframe'): ax.plot_wireframe(x, y, z)
elif (axtype == 'contour'): ax.contour3D(xgrid, ygrid, zgrid)
elif (axtype == 'contourf'): ax.contourf3D(xgrid, ygrid, zgrid)
plt.show()
def predict(network, x):
w1, w2, w3 = network['W1'], network['W2'], network['W3']
b1, b2, b3 = network['b1'], network['b2'], network['b3']
# 네트워크 구조
## 입력층: 784개 뉴런
## 첫 번째 은닉층: 50개 뉴런, sigmoid
## 두 번째 은닉층: 100개 뉴런, sigmoid
## 출력층: 10개 뉴런, softmax
a1 = np.dot(x, w1) + b1
z1 = sigmoid(a1)
a2 = np.dot(z1, w2) + b2
z2 = sigmoid(a2)
a3 = np.dot(z2, w3) + b3
y = softmax(a3)
return y
