def forward(self):
# A stable calculation, xent(sigmoid(x)) = (1 - t) + log(1 + exp(-x))
self._input[1]._result = numeric.sigmoid(self._input[0]._value)
labels = self._input[1]._value
xent = (1 - labels.astype(DTYPE)) * self._input[0]._value + numeric.log1pexp(-self._input[0]._value)
self._value = (np.sum(xent) / DTYPE(xent.size)).reshape(1)
self._gradient = np.zeros(self._value.shape, dtype=DTYPE)
def forward(self):
# A stable calculation, xent(sigmoid(x)) = (1 - t) + log(1 + exp(-x))
self._input[1]._result = numeric.sigmoid(self._input[0]._value)
labels = self._input[1]._value
xent = (1 - labels.astype(DTYPE)) * self._input[0]._value + numeric.log1pexp(-self._input[0]._value)
self._value = (np.sum(xent) / DTYPE(xent.size)).reshape(1)
self._gradient = np.zeros(self._value.shape, dtype=DTYPE)
def forward(self):
self._value = 2 * numeric.sigmoid(2 * self._input[0]._value) - 1 # tanh = 2 sig(2x) - 1
self._gradient = np.zeros(self._value.shape, dtype=DTYPE)
def forward(self):
self._value = numeric.sigmoid(self._input[0]._value)
self._gradient = np.zeros(self._value.shape, dtype=DTYPE)
def forward(self):
self._value = 2 * numeric.sigmoid(2 * self._input[0]._value) - 1 # tanh = 2 sig(2x) - 1
self._gradient = np.zeros(self._value.shape, dtype=DTYPE)
def forward(self):
self._value = numeric.sigmoid(self._input[0]._value)
self._gradient = np.zeros(self._value.shape, dtype=DTYPE)