def GRUyuce(traindata1, k, Wy, W, U, Wz, Uz, Wr, Ur, w, pianchabu):
# 初始化网络结构
uNum = k # 数据结构单元个数
hdim = k
eta = 0.1 # 学习率
#训练数据
traindata = [0 for i in range(len(traindata1))]
for i in range(len(traindata1)):
if (max(traindata1) == min(traindata1) and max(traindata1) == 0):
traindata[i] = 0.0
if (max(traindata1) == min(traindata1) and max(traindata1) != 0):
traindata[i] = 1.0
if (max(traindata1) != min(traindata1)):
traindata[i] = (traindata1[i] - min(traindata1)
) / float(max(traindata1) - min(traindata1))
#print(traindata)
# cell数据存储变量
rvalues = [[0 for col in range(hdim)] for row in range(uNum + 1)]
zvalues = [[0 for col in range(hdim)] for row in range(uNum + 1)]
hbarvalues = [[0 for col in range(hdim)] for row in range(uNum)]
hvalues = [[0 for col in range(hdim)] for row in range(uNum)]
yvalues = [0 for i in range(uNum)]
# 前向计算
rvalues[0] = function.sigmoid(function.xchenlist(traindata[0], Wr))
hbarvalues[0] = function.tanh(function.xchenlist(traindata[0], W))
zvalues[0] = function.sigmoid(function.xchenlist(traindata[0], Wz))
hvalues[0] = function.listchenlist(zvalues[0], hbarvalues[0])
yvalues[0] = function.sigmoid(function.hangchenlie(hvalues[0], Wy))
for t in range(1, uNum):
rvalues[t] = function.sigmoid(
function.listjialist(function.xchenlist(traindata[t], Wr),
function.hangchenjuzhen(hvalues[t - 1], Ur)))
hbarvalues[t] = function.tanh(
function.listjialist(
function.xchenlist(traindata[t], W),
function.hangchenjuzhen(
function.listchenlist(rvalues[t], hvalues[t - 1]), U)))
zvalues[t] = function.sigmoid(
function.listjialist(function.xchenlist(traindata[t], Wz),
function.hangchenjuzhen(hvalues[t - 1], Uz)))
hvalues[t] = function.listjialist(
function.listchenlist(function.kjianlist(1, zvalues[t]),
hvalues[t - 1]),
function.listchenlist(zvalues[t], hbarvalues[t]))
yvalues[t] = function.sigmoid(function.hangchenlie(hvalues[t], Wy))
x = function.kjialist(
min(traindata1),
function.xchenlist((max(traindata1) - min(traindata1)), yvalues))
yucezhi = function.hangchenlie(w, function.listjialist(x, pianchabu))
return yucezhi
def step(self, x, h_prev):
r = fn.sigmoid(np.dot(self.param.wr, x) + self.param.ur*h_prev + self.param.br)
z = fn.sigmoid(np.dot(self.param.wz, x) + self.param.uz*h_prev + self.param.bz)
hs = fn.tanh(np.dot(self.param.whs, x) + self.param.uhs*(r*h_prev) + self.param.bhs)
h = z*h_prev + (1-z)*hs
return r, z, hs, h
def step(self, x, h_prev):
r = fn.sigmoid(np.dot(self.param.wr, x) + self.param.ur*h_prev + np.dot(self.param.cr, self.c))
z = fn.sigmoid(np.dot(self.param.wz, x) + self.param.uz*h_prev + np.dot(self.param.cz, self.c))
hs = fn.tanh(np.dot(self.param.whs, x) + r*(self.param.uhs*h_prev + np.dot(self.param.chs, self.c)))
h = z*h_prev + (1-z)*hs
return r, z, hs, h
def GRUtrain2(traindata1, zhenshidata1, k, Wy, W, U, Wz, Uz, Wr, Ur, w):
# 初始化网络结构
uNum = k # 数据结构单元个数
xdim = 1 #输入数据维度是1
ydim = 1 #输出维度也是1
hdim = k
eta = 0.1 # 学习率
#训练数据
traindata = [[0 for col in range(len(traindata1[0]))]
for row in range(len(traindata1))]
for i in range(len(traindata1)):
for j in range(len(traindata1[0])):
if (max(traindata1[i]) == min(traindata1[i])
and max(traindata1[i]) == 0):
traindata[i][j] = 0.0
if (max(traindata1[i]) == min(traindata1[i])
and max(traindata1[i]) != 0):
traindata[i][j] = 1.0
if (max(traindata1[i]) != min(traindata1[i])):
traindata[i][j] = (traindata1[i][j] -
min(traindata1[i])) / float(
max(traindata1[i]) - min(traindata1[i]))
# print(traindata)
zhenshidata = [[0 for col in range(len(zhenshidata1[0]))]
for row in range(len(zhenshidata1))]
for i in range(len(zhenshidata1)):
for j in range(len(zhenshidata1[0])):
if (max(zhenshidata1[i]) == min(zhenshidata1[i])
and max(zhenshidata1[i]) == 0):
zhenshidata[i][j] = 0.0
if (max(zhenshidata1[i]) == min(zhenshidata1[i])
and max(zhenshidata1[i]) != 0):
zhenshidata[i][j] = 1.0
if (max(zhenshidata1[i]) != min(zhenshidata1[i])):
zhenshidata[i][j] = (
zhenshidata1[i][j] - min(zhenshidata1[i])
) / float(max(zhenshidata1[i]) - min(zhenshidata1[i]))
# cell数据存储变量
rvalues = [[0 for col in range(hdim)] for row in range(uNum + 1)]
zvalues = [[0 for col in range(hdim)] for row in range(uNum + 1)]
hbarvalues = [[0 for col in range(hdim)] for row in range(uNum)]
hvalues = [[0 for col in range(hdim)] for row in range(uNum)]
yvalues = [0 for i in range(uNum)]
#print(yvalues)
yyvalues = []
yxvalues = []
pianchaall = []
for i in range(len(traindata)):
# 前向计算
rvalues[0] = function.sigmoid(function.xchenlist(traindata[i][0], Wr))
hbarvalues[0] = function.tanh(function.xchenlist(traindata[i][0], W))
zvalues[0] = function.sigmoid(function.xchenlist(traindata[i][0], Wz))
hvalues[0] = function.listchenlist(zvalues[0], hbarvalues[0])
yvalues[0] = function.sigmoid(function.hangchenlie(hvalues[0], Wy))
for t in range(1, uNum):
rvalues[t] = function.sigmoid(
function.listjialist(
function.xchenlist(traindata[i][t], Wr),
function.hangchenjuzhen(hvalues[t - 1], Ur)))
hbarvalues[t] = function.tanh(
function.listjialist(
function.xchenlist(traindata[i][t], W),
function.hangchenjuzhen(
function.listchenlist(rvalues[t], hvalues[t - 1]), U)))
zvalues[t] = function.sigmoid(
function.listjialist(
function.xchenlist(traindata[i][t], Wz),
function.hangchenjuzhen(hvalues[t - 1], Uz)))
hvalues[t] = function.listjialist(
function.listchenlist(function.kjianlist(1, zvalues[t]),
hvalues[t - 1]),
function.listchenlist(zvalues[t], hbarvalues[t]))
yvalues[t] = function.sigmoid(function.hangchenlie(hvalues[t], Wy))
#print(yvalues)
# 反向传播
delta_r_next = [0 for i in range(hdim)]
delta_z_next = [0 for i in range(hdim)]
delta_h_next = [0 for i in range(hdim)]
delta_next = [0 for i in range(hdim)]
dWy = [0 for i in range(hdim)]
dWr = [0 for i in range(hdim)]
dUr = [[0 for col in range(hdim)] for row in range(hdim)]
dW = [0 for i in range(hdim)]
dU = [[0 for col in range(hdim)] for row in range(hdim)]
dWz = [0 for i in range(hdim)]
dUz = [[0 for col in range(hdim)] for row in range(hdim)]
#print(yvalues)
#print(uNum)
for t in range(1, uNum):
delta_y = yvalues[uNum - t] - float(zhenshidata[i][
uNum - t]) * function.sigmoidtoD(yvalues[uNum - t])
# print(delta_y)
a = function.xchenlist(delta_y, Wy)
b = function.hangchenjuzhen(delta_z_next, zip(*Uz))
c = function.listchenlist(
function.hangchenjuzhen(delta_next, zip(*U)),
rvalues[uNum - t + 1])
d = function.hangchenjuzhen(delta_r_next, zip(*Ur))
e = function.listchenlist(
delta_h_next, function.kjianlist(1, zvalues[uNum - t + 1]))
delta_h = function.listjialist(
function.listjialist(
function.listjialist(function.listjialist(a, b), c), d), e)
# print(delta_h)
delta_z = function.listchenlist(
function.listchenlist(
delta_h,
function.listjianlist(hbarvalues[uNum - t],
hvalues[uNum - t - 1])),
function.sigmoidtoD(zvalues[uNum - t]))
# print(delta_z)
delta = function.listchenlist(
function.listchenlist(delta_h, zvalues[uNum - t]),
function.tanhtoD(hbarvalues[uNum - t]))
# print(delta)
delta_r = function.listchenlist(
function.hangchenjuzhen(
function.listchenlist(
function.listchenlist(
hvalues[uNum - t - 1],
function.listchenlist(delta_h, zvalues[uNum - t])),
function.tanhtoD(hbarvalues[uNum - t])), zip(*U)),
function.sigmoidtoD(rvalues[uNum - t]))
# print(delta_r)
dWy = dWy + function.xchenlist(delta_y, hvalues[uNum - t])
dWz = dWz + function.xchenlist(traindata[i][uNum - t], delta_z)
dUz = dUz + function.liechenhang(hvalues[uNum - t - 1], delta_z)
dW = dW + function.xchenlist(traindata[i][uNum - t], delta)
dU = dU + function.liechenhang(
function.listchenlist(rvalues[uNum - t],
hvalues[uNum - t - 1]), delta)
dWr = dWr + function.xchenlist(traindata[i][uNum - t], delta_r)
dUr = dUr + function.liechenhang(hvalues[uNum - t - 1], delta_r)
delta_r_next = delta_r
delta_z_next = delta_z
delta_h_next = delta_h
delta_next = delta
t = uNum
delta_y = yvalues[uNum - t] - float(
zhenshidata[i][uNum - t]) * function.sigmoidtoD(yvalues[uNum - t])
# print(delta_y)
a = function.xchenlist(delta_y, Wy)
b = function.hangchenjuzhen(delta_z_next, zip(*Uz))
c = function.listchenlist(function.hangchenjuzhen(delta_next, zip(*U)),
rvalues[uNum - t + 1])
d = function.hangchenjuzhen(delta_r_next, zip(*Ur))
e = function.listchenlist(delta_h_next,
function.kjianlist(1, zvalues[uNum - t + 1]))
delta_h = function.listjialist(
function.listjialist(
function.listjialist(function.listjialist(a, b), c), d), e)
# print(delta_h)
delta_z = function.listchenlist(
function.listchenlist(delta_h, hbarvalues[uNum - t]),
function.sigmoidtoD(zvalues[uNum - t]))
delta = function.listchenlist(
function.listchenlist(delta_h, zvalues[uNum - t]),
function.tanhtoD(hbarvalues[uNum - t]))
delta_r = [0 for k in range(hdim)]
# print(delta)
dWy = dWy + function.xchenlist(delta_y, hvalues[uNum - t])
dWz = dWz + function.xchenlist(traindata[i][uNum - t], delta_z)
dW = dW + function.xchenlist(traindata[i][uNum - t], delta)
dWr = dWr + function.xchenlist(traindata[i][uNum - t], delta_r)
Wy = function.listjianlist(Wy, function.xchenlist(eta, dWy))
Wr = function.listjianlist(Wr, function.xchenlist(eta, dWr))
W = function.listjianlist(W, function.xchenlist(eta, dW))
Wz = function.listjianlist(Wz, function.xchenlist(eta, dWz))
Ur = function.juzhenjianjuzhen(Ur, function.xchenjuzhen(eta, dUr))
U = function.juzhenjianjuzhen(U, function.xchenjuzhen(eta, dU))
Uz = function.juzhenjianjuzhen(Uz, function.xchenjuzhen(eta, dUz))
x = function.kjialist(
min(traindata1[i]),
function.xchenlist((max(traindata1[i]) - min(traindata1[i])),
yvalues))
piancha1 = function.listjianlist(zhenshidata[i], yvalues)
piancha2 = function.listjianlist(zhenshidata1[i], x)
pianchaall.append(piancha2)
error1 = (math.sqrt(sum(function.listchenlist(piancha1,
piancha1)))) / 2.0
error2 = (math.sqrt(sum(function.listchenlist(piancha2,
piancha2)))) / 2.0
# 打印真实shuju
#print(zhenshidata[i])
# 打印预测数据
#print(yvalues)
yyvalues.append(
function.kjialist(
min(traindata1[i]),
function.xchenlist((max(traindata1[i]) - min(traindata1[i])),
yvalues)))
yxvalues.append(yvalues)
#print(yyvalues[i])
# 打印误差
#print(error1)
#print(piancha2)
#print(error2)
#print(pianchaall)
#print(yyvalues)
pianchabu = []
ss = zip(*pianchaall)
#print(ss)
for jj in range(hdim):
pianchabu.append(sum(ss[jj]) / float(len(traindata1)))
#print(pianchabu)
#print(yxvalues)
#print(zhenshidata)
#print(zhenshidata[0][hdim-1])
for j in range(hdim):
x = 0
for kk in range(len(yyvalues)):
x = x + (zhenshidata[kk][hdim - 1] -
function.hangchenlie(w, yxvalues[kk])) * yxvalues[kk][j]
w[j] = w[j] + eta * x
#print(w)
return Wy, W, U, Wz, Uz, Wr, Ur, w, pianchabu, error2