def __init__(self,
kernel_shape=list,
input_variable=jen.Variable,
name=str,
stride=1,
padding='SAME'):
# kernel_shape = [ksize, ksize, input_channels, output_channels]
for i in kernel_shape:
if not isinstance(i, int):
raise Exception(
"Operator Conv2D name: %s kernel shape is not list of int"
% self.name)
if not isinstance(input_variable, jen.Variable):
raise Exception(
"Operator Conv2D name: %s's input_variable is not instance of Variable"
% name)
if len(input_variable.shape) != 4:
raise Exception(
"Operator Conv2D name: %s's input_variable's shape != 4d Variable!"
% name)
self.ksize = kernel_shape[0]
self.stride = stride
self.output_num = kernel_shape[-1]
self.padding = padding
self.col_image = []
self.weights = jen.Variable(kernel_shape,
scope=name,
name='weights',
learnable=True)
self.bias = jen.Variable([self.output_num],
scope=name,
name='bias',
learnable=True)
self.batch_size = input_variable.shape[0]
if self.padding == 'SAME':
_output_shape = [
self.batch_size,
int(input_variable.shape[1] / stride),
int(input_variable.shape[2] / stride), self.output_num
]
if self.padding == 'VALID':
_output_shape = [
self.batch_size,
int((input_variable.shape[1] - self.ksize + 1) / stride),
int((input_variable.shape[2] - self.ksize + 1) / stride),
self.output_num
]
self.output_variables = jen.Variable(_output_shape,
name='out',
scope=name) # .name
self.input_variables = input_variable
super(Conv2D, self).__init__(name, self.input_variables,
self.output_variables)
def __init__(self, predict=jen.Variable, label=jen.Variable, name=str):
self.batch_size = predict.shape[0]
self.input_variables = [predict, label]
self.loss = jen.Variable([1], name='loss', scope=name)
self.prediction = jen.Variable(predict.shape,
name='prediction',
scope=name)
self.softmax = np.zeros(self.prediction.shape)
self.output_variables = [self.loss, self.prediction]
super(SoftmaxLoss, self).__init__(name, self.input_variables,
self.output_variables)
def __init__(self, input_variable=jen.Variable, name=str):
self.input_variables = input_variable
self.output_variables = jen.Variable(self.input_variables.shape,
name='out',
scope=name)
super(Tanh, self).__init__(name, self.input_variables,
self.output_variables)
def __init__(self,
ksize=2,
input_variable=jen.Variable,
name=str,
stride=2):
if not isinstance(input_variable, jen.Variable):
raise Exception(
"Operator Conv2D name: %s's input_variable is not instance of Variable"
% name)
self.ksize = ksize
self.stride = stride
self.batch_size = input_variable.shape[0]
self.output_channels = input_variable.shape[-1]
self.index = np.zeros(input_variable.shape)
self.input_variables = input_variable
_output_shape = [
self.batch_size,
int(input_variable.shape[2] / stride),
int(input_variable.shape[2] / stride), self.output_channels
]
self.output_variables = jen.Variable(_output_shape,
name='out',
scope=name)
super(MaxPooling, self).__init__(name, self.input_variables,
self.output_variables)
def __init__(self, input_variable=jen.Variable, name=str, alpha=0.1):
self.input_variables = input_variable
self.output_variables = jen.Variable(self.input_variables.shape,
name='out',
scope=name)
self.alpha = alpha
super(Elu, self).__init__(name, self.input_variables,
self.output_variables)
def __init__(self, input_variable=jen.Variable, name=str, alpha=0.25):
self.input_variables = input_variable
self.output_variables = jen.Variable(self.input_variables.shape,
name='out',
scope=name)
self.alpha = alpha
self.momentum = 0.9
self.eta = 1e-4
super(Prelu, self).__init__(name, self.input_variables,
self.output_variables)
def __init__(self, name, phase, input_variable=jen.Variable, prob=0.5):
self.input_variables = input_variable
self.output_variables = jen.Variable(shape=input_variable.shape,
scope=name,
name='out')
self.prob = prob
self.phase = phase
self.index = np.ones(input_variable.shape)
super(DropOut, self).__init__(name, self.input_variables,
self.output_variables)
def __init__(self, output_num, input_variable=jen.Variable, name=str):
if not isinstance(input_variable, jen.Variable):
raise Exception(
"Operator Conv2D name: %s's input_variable is not instance of Variable"
% name)
self.batch_size = input_variable.shape[0]
input_len = ft.reduce(lambda x, y: x * y, input_variable.shape[1:])
self.output_num = output_num
self.weights = jen.Variable([input_len, self.output_num],
name='weights',
scope=name,
learnable=True)
self.bias = jen.Variable([self.output_num],
name='bias',
scope=name,
learnable=True)
self.output_variables = jen.Variable(
[self.batch_size, self.output_num], name='out', scope=name)
self.input_variables = input_variable
super(FullyConnect, self).__init__(name, self.input_variables,
self.output_variables)