def clip(a, a_min, a_max):
"""Clips the values of an array to a given interval.
Given an interval, values outside the interval are clipped to the
interval edges. For example, if an interval of ``[0, 1]`` is specified,
values smaller than 0 become 0, and values larger than 1 become 1.
Args:
a (~chainerx.ndarray): Array containing elements to clip.
a_min (scalar): Maximum value.
a_max (scalar): Minimum value.
Returns:
~chainerx.ndarray: An array with the elements of ``a``, but where
values < ``a_min`` are replaced with ``a_min``,
and those > ``a_max`` with ``a_max``.
Note:
The :class:`~chainerx.ndarray` typed ``a_min`` and ``a_max`` are
not supported yet.
Note:
During backpropagation, this function propagates the gradient
of the output array to the input array ``a``.
.. seealso:: :func:`numpy.clip`
"""
return -chainerx.maximum(-chainerx.maximum(a, a_min), -a_max)
def test_maximum_invalid_dtypes(device, dtype):
shape = (3, 2)
bool_array = chainerx.array(array_utils.uniform(shape, 'bool_'))
numeric_array = chainerx.array(array_utils.uniform(shape, dtype))
with pytest.raises(chainerx.DtypeError):
chainerx.maximum(bool_array, numeric_array)
with pytest.raises(chainerx.DtypeError):
chainerx.maximum(numeric_array, bool_array)
def __call__(self, x):
h = self.bn1(self.conv1(x))
h = chx.max_pool(chx.maximum(0, h), 3, stride=2)
h = self.res2(h)
h = self.res3(h)
h = self.res4(h)
h = self.res5(h)
h = chx.average_pool(h, 7, stride=1)
h = self.fc(h)
return h
def __call__(self, x):
h = self.bn1(self.conv1(x))
h = chx.max_pool(chx.maximum(0, h), 3, stride=2)
h = self.res2(h)
h = self.res3(h)
h = self.res4(h)
h = self.res5(h)
h = chx.average_pool(h, 7, stride=1)
h = self.fc(h)
return h
def forward_chainerx(self, inputs):
x, = inputs
y = chainerx.max_pool(
x, ksize=self.ksize, stride=self.stride, pad=self.pad,
cover_all=self.cover_all)
# This function can return -inf (or huge negative numbers in case of
# CUDA) around boundaries.
# Convert them to finite numbers in order to properly calculate numeric
# gradients.
y = chainerx.maximum(y, -1e4)
return y,
def forward_chainerx(self, inputs):
x, = inputs
y = chainerx.max_pool(x,
ksize=self.ksize,
stride=self.stride,
pad=self.pad,
cover_all=self.cover_all)
# This function can return -inf (or huge negative numbers in case of
# CUDA) around boundaries.
# Convert them to finite numbers in order to properly calculate numeric
# gradients.
y = chainerx.maximum(y, -1e4)
return y,
def relu(x):
"""Rectified Linear Unit function.
Args:
x (~chainerx.ndarray): Input array.
Returns:
:class:`~chainerx.ndarray`: Returned array: :math:`y = \\max (0, x)`.
Note:
During backpropagation, this function propagates the gradient of the
output array to the input array ``x``.
"""
# TODO(imanishi): The function should also be available to C++ users
return chainerx.maximum(0, x)
def relu(x):
"""Rectified Linear Unit function.
Args:
x (~chainerx.ndarray): Input array.
Returns:
:class:`~chainerx.ndarray`: Returned array: :math:`y = \\max (0, x)`.
Note:
During backpropagation, this function propagates the gradient of the
output array to the input array ``x``.
"""
# TODO(imanishi): The function should also be available to C++ users
return chainerx.maximum(0, x)
def forward_chainerx(self, inputs):
x, = inputs
return chainerx.maximum(x, 0),
def forward_chainerx(self, inputs):
x, = inputs
return chainerx.maximum(x, 0),
def __call__(self, x):
h = chx.maximum(0, self.bn1(self.conv1(x)))
h = chx.maximum(0, self.bn2(self.conv2(h)))
h = self.bn3(self.conv3(h))
return chx.maximum(0, h + x)
def __call__(self, x):
h1 = chx.maximum(0, self.bn1(self.conv1(x)))
h1 = chx.maximum(0, self.bn2(self.conv2(h1)))
h1 = self.bn3(self.conv3(h1))
h2 = self.bn4(self.conv4(x))
return chx.maximum(0, h1 + h2)
def forward_chainerx(self, inputs):
x, = inputs
return chainerx.minimum(chainerx.maximum(0, x), self.cap),
def forward_chainerx(self, inputs):
x, = inputs
return chainerx.minimum(chainerx.maximum(0, x), self.cap),
def __call__(self, x):
h = chx.maximum(0, self.bn1(self.conv1(x)))
h = chx.maximum(0, self.bn2(self.conv2(h)))
h = self.bn3(self.conv3(h))
return chx.maximum(0, h + x)
def __call__(self, x):
h1 = chx.maximum(0, self.bn1(self.conv1(x)))
h1 = chx.maximum(0, self.bn2(self.conv2(h1)))
h1 = self.bn3(self.conv3(h1))
h2 = self.bn4(self.conv4(x))
return chx.maximum(0, h1 + h2)
