def model(dr):
num_input = dr.num_feature
# num_hidden1 = 32
# num_hidden2 = 16
# num_hidden3 = 8
# num_hidden4 = 4
nums_hidden = [64, 32, 16, 8, 4]
num_output = 1
max_epoch = 2000
batch_size = 16
learning_rate = 0.1
params = HyperParameters_4_0(learning_rate,
max_epoch,
batch_size,
net_type=NetType.Fitting,
init_method=InitialMethod.MSRA,
stopper=Stopper(StopCondition.StopDiff, 1e-7))
net = NeuralNet_4_0(params, "HouseSingle")
fc1 = FcLayer_1_0(num_input, nums_hidden[0], params)
net.add_layer(fc1, "fc1")
r1 = ActivationLayer(Relu())
net.add_layer(r1, "r1")
fc2 = FcLayer_1_0(nums_hidden[0], nums_hidden[1], params)
net.add_layer(fc2, "fc2")
r2 = ActivationLayer(Relu())
net.add_layer(r2, "r2")
fc3 = FcLayer_1_0(nums_hidden[1], nums_hidden[2], params)
net.add_layer(fc3, "fc3")
r3 = ActivationLayer(Relu())
net.add_layer(r3, "r3")
fc4 = FcLayer_1_0(nums_hidden[2], nums_hidden[3], params)
net.add_layer(fc4, "fc4")
r4 = ActivationLayer(Relu())
net.add_layer(r4, "r4")
if len(nums_hidden) == 4:
fc5 = FcLayer_1_0(nums_hidden[3], num_output, params)
net.add_layer(fc5, "fc5")
if len(nums_hidden) == 5:
fc5 = FcLayer_1_0(nums_hidden[3], nums_hidden[4], params)
net.add_layer(fc5, "fc5")
r5 = ActivationLayer(Relu())
net.add_layer(r5, "r5")
fc6 = FcLayer_1_0(nums_hidden[4], num_output, params)
net.add_layer(fc6, "fc6")
net.train(dr, checkpoint=10, need_test=True)
return net
def model():
dr = LoadData()
num_input = dr.num_feature
num_hidden1 = 64
num_hidden2 = 64
num_hidden3 = 32
num_hidden4 = 16
num_output = 1
max_epoch = 100
batch_size = 16
learning_rate = 0.1
params = HyperParameters_4_0(learning_rate,
max_epoch,
batch_size,
net_type=NetType.BinaryClassifier,
init_method=InitialMethod.Xavier,
stopper=Stopper(StopCondition.StopDiff, 1e-3))
net = NeuralNet_4_0(params, "Income")
fc1 = FcLayer_1_0(num_input, num_hidden1, params)
net.add_layer(fc1, "fc1")
a1 = ActivationLayer(Relu())
net.add_layer(a1, "relu1")
fc2 = FcLayer_1_0(num_hidden1, num_hidden2, params)
net.add_layer(fc2, "fc2")
a2 = ActivationLayer(Relu())
net.add_layer(a2, "relu2")
fc3 = FcLayer_1_0(num_hidden2, num_hidden3, params)
net.add_layer(fc3, "fc3")
a3 = ActivationLayer(Relu())
net.add_layer(a3, "relu3")
fc4 = FcLayer_1_0(num_hidden3, num_hidden4, params)
net.add_layer(fc4, "fc4")
a4 = ActivationLayer(Relu())
net.add_layer(a4, "relu4")
fc5 = FcLayer_1_0(num_hidden4, num_output, params)
net.add_layer(fc5, "fc5")
logistic = ClassificationLayer(Logistic())
net.add_layer(logistic, "logistic")
#net.load_parameters()
net.train(dr, checkpoint=10, need_test=True)
net.ShowLossHistory()
def model():
dataReader = LoadData()
num_input = 1
num_hidden1 = 4
num_output = 1
max_epoch = 10000
batch_size = 10
learning_rate = 0.5
eps = 1e-5
params = HyperParameters_4_0(learning_rate,
max_epoch,
batch_size,
net_type=NetType.Fitting,
init_method=InitialMethod.Xavier,
stopper=Stopper(StopCondition.StopLoss,
0.001))
net = NeuralNet_4_0(params, "Level1_CurveFittingNet")
fc1 = FcLayer_1_0(num_input, num_hidden1, params)
net.add_layer(fc1, "fc1")
sigmoid1 = ActivationLayer(Sigmoid())
net.add_layer(sigmoid1, "sigmoid1")
fc2 = FcLayer_1_0(num_hidden1, num_output, params)
net.add_layer(fc2, "fc2")
net.train(dataReader, checkpoint=100, need_test=True)
net.ShowLossHistory()
ShowResult(net, dataReader)
def model(dataReader):
num_input = 2
num_hidden = 3
num_output = 1
max_epoch = 1000
batch_size = 5
learning_rate = 0.1
params = HyperParameters_4_1(learning_rate,
max_epoch,
batch_size,
net_type=NetType.BinaryClassifier,
init_method=InitialMethod.Xavier,
stopper=Stopper(StopCondition.StopLoss, 0.02))
net = NeuralNet_4_1(params, "Arc")
fc1 = FcLayer_1_1(num_input, num_hidden, params)
net.add_layer(fc1, "fc1")
sigmoid1 = ActivationLayer(Sigmoid())
net.add_layer(sigmoid1, "sigmoid1")
fc2 = FcLayer_1_1(num_hidden, num_output, params)
net.add_layer(fc2, "fc2")
logistic = ClassificationLayer(Logistic())
net.add_layer(logistic, "logistic")
net.train(dataReader, checkpoint=10, need_test=True)
return net
def model(dr, lr, bs):
num_input = dr.num_feature
num_hidden1 = 32
num_hidden2 = 16
num_hidden3 = 8
num_hidden4 = 4
num_output = 1
max_epoch = 1000
#batch_size = 16
batch_size = bs
learning_rate = lr
params = HyperParameters_4_0(learning_rate,
max_epoch,
batch_size,
net_type=NetType.Fitting,
init_method=InitialMethod.MSRA,
stopper=Stopper(StopCondition.StopDiff, 1e-7))
net = NeuralNet_4_0(params, "HouseSingle")
fc1 = FcLayer_1_0(num_input, num_hidden1, params)
net.add_layer(fc1, "fc1")
r1 = ActivationLayer(Relu())
net.add_layer(r1, "r1")
fc2 = FcLayer_1_0(num_hidden1, num_hidden2, params)
net.add_layer(fc2, "fc2")
r2 = ActivationLayer(Relu())
net.add_layer(r2, "r2")
fc3 = FcLayer_1_0(num_hidden2, num_hidden3, params)
net.add_layer(fc3, "fc3")
r3 = ActivationLayer(Relu())
net.add_layer(r3, "r3")
fc4 = FcLayer_1_0(num_hidden3, num_hidden4, params)
net.add_layer(fc4, "fc4")
r4 = ActivationLayer(Relu())
net.add_layer(r4, "r4")
fc5 = FcLayer_1_0(num_hidden4, num_output, params)
net.add_layer(fc5, "fc5")
net.train(dr, checkpoint=10, need_test=True)
return net
def net(init_method, activator):
max_epoch = 1
batch_size = 5
learning_rate = 0.02
eps = 0.01
params = HyperParameters_4_1(learning_rate,
max_epoch,
batch_size,
eps,
net_type=NetType.Fitting,
init_method=init_method)
net = NeuralNet_4_1(params, "level1")
num_hidden = [128, 128, 128, 128, 128, 128, 128]
fc_count = len(num_hidden) - 1
layers = []
for i in range(fc_count):
fc = FcLayer_1_1(num_hidden[i], num_hidden[i + 1], params)
net.add_layer(fc, "fc")
layers.append(fc)
ac = ActivationLayer(activator)
net.add_layer(ac, "activator")
layers.append(ac)
# end for
# 从正态分布中取1000个样本，每个样本有num_hidden[0]个特征值
# 转置是为了可以和w1做矩阵乘法
x = np.random.randn(1000, num_hidden[0])
# 激活函数输出值矩阵列表
a_value = []
# 依次做所有层的前向计算
input = x
for i in range(len(layers)):
output = layers[i].forward(input)
# 但是只记录激活层的输出
if isinstance(layers[i], ActivationLayer):
a_value.append(output)
# end if
input = output
# end for
for i in range(len(a_value)):
ax = plt.subplot(1, fc_count + 1, i + 1)
ax.set_title("layer" + str(i + 1))
plt.ylim(0, 10000)
if i > 0:
plt.yticks([])
ax.hist(a_value[i].flatten(), bins=25, range=[0, 1])
#end for
# super title
plt.suptitle(init_method.name + " : " + activator.get_name())
plt.show()
def __init__(self, init_method, activator, stop_thresh=0.12):
#固定超参数
self.dataReader = LoadData()
self.num_input = self.dataReader.num_feature
self.num_hidden1 = 128
self.num_hidden2 = 64
self.num_hidden3 = 32
self.num_hidden4 = 16
self.num_output = 10
self.max_epoch = 20
self.batch_size = 64
self.learning_rate = 0.1
params = HyperParameters_4_1(self.learning_rate,
self.max_epoch,
self.batch_size,
net_type=NetType.MultipleClassifier,
init_method=init_method,
stopper=Stopper(StopCondition.Nothing, 0))
net = NeuralNet_4_1(params, "MNIST")
fc1 = FcLayer_1_1(self.num_input, self.num_hidden1, params)
net.add_layer(fc1, "fc1")
r1 = ActivationLayer(Relu())
net.add_layer(r1, "r1")
fc2 = FcLayer_1_1(self.num_hidden1, self.num_hidden2, params)
net.add_layer(fc2, "fc2")
r2 = ActivationLayer(Relu())
net.add_layer(r2, "r2")
fc3 = FcLayer_1_1(self.num_hidden2, self.num_hidden3, params)
net.add_layer(fc3, "fc3")
r3 = ActivationLayer(Relu())
net.add_layer(r3, "r3")
fc4 = FcLayer_1_1(self.num_hidden3, self.num_hidden4, params)
net.add_layer(fc4, "fc4")
r4 = ActivationLayer(Relu())
net.add_layer(r4, "r4")
fc5 = FcLayer_1_1(self.num_hidden4, self.num_output, params)
net.add_layer(fc5, "fc5")
softmax = ClassificationLayer(Softmax())
net.add_layer(softmax, "softmax")
self.net = net
def model_relu(num_input, num_hidden, num_output, hp):
net = NeuralNet_4_1(hp, "chinabank_relu")
fc1 = FcLayer_1_1(num_input, num_hidden, hp)
net.add_layer(fc1, "fc1")
r1 = ActivationLayer(Relu())
net.add_layer(r1, "Relu1")
fc2 = FcLayer_1_1(num_hidden, num_hidden, hp)
net.add_layer(fc2, "fc2")
r2 = ActivationLayer(Relu())
net.add_layer(r2, "Relu2")
fc3 = FcLayer_1_1(num_hidden, num_output, hp)
net.add_layer(fc3, "fc3")
softmax = ClassificationLayer(Softmax())
net.add_layer(softmax, "softmax")
net.train(dataReader, checkpoint=50, need_test=True)
net.ShowLossHistory()
ShowResult(net, hp.toString())
ShowData(dataReader)
def model_sigmoid(num_input, num_hidden, num_output, hp):
net = NeuralNet_4_1(hp, "chinabank_sigmoid")
fc1 = FcLayer_1_1(num_input, num_hidden, hp)
net.add_layer(fc1, "fc1")
s1 = ActivationLayer(Sigmoid())
net.add_layer(s1, "Sigmoid1")
fc2 = FcLayer_1_1(num_hidden, num_output, hp)
net.add_layer(fc2, "fc2")
softmax1 = ClassificationLayer(Softmax())
net.add_layer(softmax1, "softmax1")
net.train(dataReader, checkpoint=50, need_test=True)
net.ShowLossHistory()
ShowResult(net, hp.toString())
ShowData(dataReader)
max_epoch = 30
batch_size = 64
learning_rate = 0.1
params = HyperParameters_4_0(learning_rate,
max_epoch,
batch_size,
net_type=NetType.MultipleClassifier,
init_method=InitialMethod.MSRA,
stopper=Stopper(StopCondition.StopLoss, 0.26))
net = NeuralNet_4_0(params, "Cifar10")
fc1 = FcLayer_1_0(num_input, num_hidden1, params)
net.add_layer(fc1, "fc1")
r1 = ActivationLayer(Relu())
net.add_layer(r1, "r1")
fc2 = FcLayer_1_0(num_hidden1, num_hidden2, params)
net.add_layer(fc2, "fc2")
r2 = ActivationLayer(Relu())
net.add_layer(r2, "r2")
fc3 = FcLayer_1_0(num_hidden2, num_hidden3, params)
net.add_layer(fc3, "fc3")
r3 = ActivationLayer(Relu())
net.add_layer(r3, "r3")
fc4 = FcLayer_1_0(num_hidden3, num_hidden4, params)
net.add_layer(fc4, "fc4")
r4 = ActivationLayer(Relu())
def model():
dr = LoadData()
num_input = dr.num_feature
num_hidden1 = 32
num_hidden2 = 16
num_hidden3 = 8
num_hidden4 = 4
num_output = 1
max_epoch = 1000
batch_size = 16
learning_rate = 0.01
params = HyperParameters_4_0(learning_rate,
max_epoch,
batch_size,
net_type=NetType.Fitting,
init_method=InitialMethod.Xavier,
stopper=Stopper(StopCondition.StopDiff, 1e-6))
net = NeuralNet_4_0(params, "HouseSingle")
fc1 = FcLayer_1_0(num_input, num_hidden1, params)
net.add_layer(fc1, "fc1")
r1 = ActivationLayer(Relu())
net.add_layer(r1, "r1")
fc2 = FcLayer_1_0(num_hidden1, num_hidden2, params)
net.add_layer(fc2, "fc2")
r2 = ActivationLayer(Relu())
net.add_layer(r2, "r2")
fc3 = FcLayer_1_0(num_hidden2, num_hidden3, params)
net.add_layer(fc3, "fc3")
r3 = ActivationLayer(Relu())
net.add_layer(r3, "r3")
fc4 = FcLayer_1_0(num_hidden3, num_hidden4, params)
net.add_layer(fc4, "fc4")
r4 = ActivationLayer(Relu())
net.add_layer(r4, "r4")
fc5 = FcLayer_1_0(num_hidden4, num_output, params)
net.add_layer(fc5, "fc5")
#ShowResult(net, dr)
#net.load_parameters()
#Inference(net, dr)
#exit()
#ShowResult(net, dr)
net.train(dr, checkpoint=10, need_test=True)
output = net.inference(dr.XTest)
real_output = dr.DeNormalizeY(output)
mse = np.sum((dr.YTestRaw - real_output)**2) / dr.YTest.shape[0] / 10000
print("mse=", mse)
net.ShowLossHistory()
ShowResult(net, dr)
batch_size = 128
learning_rate = 0.1
eta = 0.2
eps = 0.01
params = HyperParameters_4_0(learning_rate,
max_epoch,
batch_size,
net_type=NetType.Fitting,
init_method=InitialMethod.Xavier,
stopper=Stopper(StopCondition.StopDiff, 1e-7))
net = NeuralNet_4_0(params, "SAFE")
fc1 = FcLayer_1_0(num_input, num_hidden1, params)
net.add_layer(fc1, "fc1")
r1 = ActivationLayer(Sigmoid())
net.add_layer(r1, "r1")
fc2 = FcLayer_1_0(num_hidden1, num_hidden2, params)
net.add_layer(fc2, "fc2")
r2 = ActivationLayer(Sigmoid())
net.add_layer(r2, "r2")
fc3 = FcLayer_1_0(num_hidden2, num_output, params)
net.add_layer(fc3, "fc3")
r3 = ActivationLayer(Sigmoid())
net.add_layer(r3, "r3")
softmax = ClassificationLayer(Softmax())
net.add_layer(softmax, "softmax")
num_output = 1
max_epoch = 10000
batch_size = 5
learning_rate = 0.1
params = HyperParameters_4_0(learning_rate,
max_epoch,
batch_size,
net_type=NetType.BinaryClassifier,
init_method=InitialMethod.Xavier,
stopper=Stopper(StopCondition.StopLoss, 0.05))
net = NeuralNet_4_0(params, "Arc")
fc1 = FcLayer_1_0(num_input, num_hidden, params)
net.add_layer(fc1, "fc1")
sigmoid1 = ActivationLayer(Sigmoid())
net.add_layer(sigmoid1, "sigmoid1")
fc2 = FcLayer_1_0(num_hidden, num_output, params)
net.add_layer(fc2, "fc2")
logistic = ClassificationLayer(Logistic())
net.add_layer(logistic, "logistic")
#net.load_parameters()
net.train(dataReader, checkpoint=10, need_test=True)
net.ShowLossHistory()
ShowResult2D(net, dataReader)