def __call__(self, *args, **kwargs):
arg0 = args[0]
out = self.__creator__(-arg0.value)
out.graph = GraphPartial([arg0], self.name, out, lambda g, a0: (-g, ))
return out
def __call__(self, *args, **kwargs):
arg0 = args[0]
out = self.__creator__(np.log(arg0.value))
out.graph = GraphPartial([arg0], self.name, out, self.grad_fn)
return out
def __call__(self, *args, **kwargs):
arg0 = args[0]
out = self.__creator__(np.exp(arg0.value))
out.graph = GraphPartial([arg0], self.name, out,
lambda g, a0: self.grad_fn(g, a0, out.value))
return out
def __call__(self, *args, **kwargs):
arg0 = args[0]
v = arg0.value
out = self.__creator__(1 / (1 + np.exp(-v)))
out.graph = GraphPartial([arg0], self.name, out, self.grad_fn)
return out
def __call__(self, *args, **kwargs):
arg0 = args[0]
arg1 = args[1]
out = self.__creator__(np.matmul(arg0.value, arg1.value))
out.graph = GraphPartial([arg0, arg1], self.name, out, self.grad_fn)
return out
def __call__(self, *args, **kwargs):
arg0 = args[0]
out = self.__creator__(np.sqrt(arg0.value))
out.graph = GraphPartial([arg0], self.name, out, lambda g, a0:
(g * 0.5 * a0**-0.5))
return out
def __call__(self, *args, **kwargs):
arg0 = args[0]
arg1 = args[1]
out = self.__creator__(arg0.value - arg1.value)
out.graph = GraphPartial([arg0, arg1], self.name, out,
lambda g, a0, a1: (g, -g))
return out
def __call__(self, *args, **kwargs):
arg0 = args[0]
axis = kwargs.get('axis', None)
out = self.__creator__(np.transpose(arg0.value, axis))
out.graph = GraphPartial([arg0], self.name, out,
lambda g, a0: self.grad_fn(g, a0, axis))
return out
def __call__(self, *args, **kwargs):
arg0 = args[0]
axis = kwargs.get('axis', 0) or 0
out = self.__creator__(arg0.value.sum(axis))
out.graph = GraphPartial(
[arg0], self.name, out,
lambda g, a0: self.grad_fn(g, a0, arg0.value.shape[axis]))
return out
def __call__(self, *args, **kwargs):
arg0 = args[0]
v = arg0.value
# mask =np.ma.masked_less_equal(v, 0)
mask = np.ma.masked_greater(v, 0)
v[v <= 0] = 0
out = self.__creator__(v)
out.graph = GraphPartial([arg0], self.name, out,
lambda g, a0: self.grad_fn(g, a0, mask))
return out
def __call__(self, *args, **kwargs):
arg0 = args[0]
idx = args[1]
out = self.__creator__(arg0.value[idx])
if isinstance(idx, list) and (len(idx) == 0
or not isinstance(idx[0], slice)):
idx = np.array(idx, dtype='int64')
out.graph = GraphPartial([arg0], self.name, out,
lambda g, a0: self.grad_fn(g, a0, idx))
return out