def get_net(self, data_layer):
n = Network()
n.add(data_layer)
n.name = "LogisticRegressionNet"
l = InnerProduct(name="fc1", top="fc1", bottom="data", num_output=2)
l3 = Loss(name="loss", type=LossType.SoftmaxWithLoss, bottom=["fc1", "label"], include=Phase.TRAIN)
l4 = Loss(name="output", type= LossType.SOFTMAX, top= "output", bottom="fc1", include=Phase.TEST)
n.add(l)
n.add(l3)
n.add(l4)
return n.str().__str__()
def get_net(self, data_layer):
n = Network()
n.add(data_layer)
n.name = "LogisticRegressionNet"
l = InnerProduct(name="fc1", top="fc1", bottom="data", num_output=2)
l3 = Loss(name="loss",
type=LossType.SoftmaxWithLoss,
bottom=["fc1", "label"],
include=Phase.TRAIN)
l4 = Loss(name="output",
type=LossType.SOFTMAX,
top="output",
bottom="fc1",
include=Phase.TEST)
n.add(l)
n.add(l3)
n.add(l4)
return n.str().__str__()
def get_net(self, data_layer):
net = Network()
net.add(data_layer)
net.name = "LeNet"
c_bottom = ['data', 'pool0']
c_num_output = [20, 50]
max= pb2.PoolingParameter.MAX
xavier = pb2.FillerParameter(type= 'xavier')
const = pb2.FillerParameter(type= 'constant')
for i in range(0,2):
c = 'conv%d'%i
p = "pool%d"%i
r = "relu%d"%i
n = "norm%d"%i
c_layer = Convolution(name=c , bottom= c_bottom[i] , num_output = c_num_output[i] , kernel_size =5, stride=1, weight_filler= xavier , bias_filler = const)
p_layer = Pooling(name= p, bottom= c, kernel_size=2 , stride =2, pool = max )
net.add(c_layer), net.add(p_layer),
ip1 = InnerProduct(name="ip1", bottom="pool1", num_output=500, weight_filler= xavier , bias_filler = const)
r_layer = Relu(name = r, bottom= "ip1", top= "ip1")
ip2 = InnerProduct(name="ip2", bottom="ip1", num_output=10, weight_filler= xavier , bias_filler = const)
o1 = Loss(name="loss", type=LossType.SoftmaxWithLoss, bottom=["ip2", "label"], include=Phase.TRAIN)
o2 = Loss(name="output", type= LossType.SOFTMAX, top= "output", bottom="ip2", include=Phase.TEST)
net.add(ip1), net.add(r_layer), net.add(ip2), net.add(o1), net.add(o2)
return net.str().__str__()