class Relu(Function):
Input('in_array', (chainer.Variable,))
Output('out_array', (chainer.Variable,))
def call(self):
return self.ID + ' = relu('
@classmethod
def register_chainer_impl(cls):
return chainer.functions.relu
class HardSigmoid(Function):
Input('in_array', (chainer.Variable,))
Output('out_array', (chainer.Variable,))
def call(self):
return self.ID + ' = hard_sigmoid('
@classmethod
def register_chainer_impl(cls):
return chainer.functions.hard_sigmoid
class LeakyRelu(Function):
Input('in_array', (chainer.Variable,))
Input('slope', (float,))
Output('out_array', (chainer.Variable,))
def call(self):
return self.ID + ' = leaky_relu(slope={0}, x='.format(self._slope)
@classmethod
def register_chainer_impl(cls):
return chainer.functions.leaky_relu
class ClippedRelu(Function):
Input('in_array', (chainer.Variable,))
Input('z', (float,))
Output('out_array', (chainer.Variable,))
def call(self):
self.check_member(('_alpha',))
return self.ID + ' = clipped_relu(z={0}, x='.format(self._z)
@classmethod
def register_chainer_impl(cls):
return chainer.functions.clipped_relu
class Dropout(Function):
Input('in_array', (chainer.Variable,))
Input('ratio', (float,))
Output('out_array', (chainer.Variable,))
def call(self):
self.check_member(('_ratio',))
return self.ID + ' = dropout(ratio={0}, x='.format(self._ratio)
@classmethod
def register_chainer_impl(cls):
return chainer.functions.dropout
class VGG16(Link):
Input('in_array', (chainer.Variable, ))
Input('pretrained_model', (str, ))
Output('out_array', (chainer.Variable, ))
is_image_node = True
def call_init(self):
return 'VGG16Layers({pretrained_model}),' \
.format(pretrained_model=self._pretrained_model)
@classmethod
def register_chainer_impl(cls):
return chainer.links.VGG16Layers
class AveragePooling2d(Function):
Input('in_array', (chainer.Variable, ))
Input('ksize', (int, ))
Input('pad', (int, ))
Output('out_array', (chainer.Variable, ))
is_image_node = True
def call(self):
return self.ID + ' = average_pooling_2d(ksize={0}, pad={1}, x='.\
format(self._ksize, self._pad)
@classmethod
def register_chainer_impl(cls):
return chainer.functions.average_pooling_2d
class LocalResponseNormalization(Function):
Input('in_array', (chainer.Variable, ))
Input('n', (int, ))
Input('k', (int, ))
Input('alpha', (float, ))
Input('beta', (float, ))
Output('out_array', (chainer.Variable, ))
is_image_node = True
def call(self):
return self.ID + ' = local_response_normalization(n={n}, k={k}, x=' \
.format(n=self._n,
k=self._k)
@classmethod
def register_chainer_impl(cls):
return chainer.functions.LocalResponseNormalization
class BatchNormalization(Link):
Input('in_array', (chainer.Variable, ))
Input('size', (int, ))
Input('decay', (float, ))
Input('eps', (float, ))
Input('use_gamma', (bool, ), select=[True, False])
Input('use_beta', (bool, ), select=[True, False])
Output('out_array', (chainer.Variable, ))
is_image_node = True
def call_init(self):
return 'BatchNormalization({size}, {decay}, {eps}, numpy.float32, ' \
'{use_gamma}, {use_beta}),' \
.format(size=self._size,
decay=self._decay,
eps=self._eps,
use_gamma=self._use_gamma,
use_beta=self._use_beta)
@classmethod
def register_chainer_impl(cls):
return chainer.links.BatchNormalization
class Convolution2D(Link):
Input('in_array', (chainer.Variable, ))
Input('out_channels', (int, ))
Input('ksize', (int, ))
Input('stride', (int, ))
Input('pad', (int, ))
Input('nobias', (bool, ), select=[True, False])
Output('out_array', (chainer.Variable, ))
is_image_node = True
def call_init(self):
return 'Convolution2D({in_channels}, {out_channels}, {ksize}, ' \
'{stride}, {pad}, {nobias}),' \
.format(in_channels='None',
out_channels=self._out_channels,
ksize=self._ksize,
stride=self._stride,
pad=self._pad,
nobias=self._nobias)
@classmethod
def register_chainer_impl(cls):
return chainer.links.Convolution2D