def _oper_gpu(cls, lhs, rhs):
log_lhs = log(lhs + 1e-8)
ret = cls._create_node(-cu.cusum(get_gpu(log_lhs * rhs)))
ret.attrs._log_lhs = log_lhs
ret.attrs._rhs = rhs
ret.attrs._lhs = lhs
return ret
def _oper_gpu(cls, lhs, rhs, clip):
N = len(lhs)
value = cu.cusum(get_gpu((get_gpu(lhs) - get_gpu(rhs)) ** 2)) / (N * 2)
ret = cls._create_node(value)
ret.attrs._lhs = lhs
ret.attrs._rhs = rhs
ret.attrs._clip = clip
return ret
def _oper_gpu(cls, lhs, rhs):
N = lhs.shape[0]
z = softmax(lhs)
tmp1 = get_gpu(lhs).empty_like_me()
cu.cucross_entropy(get_gpu(z), get_gpu(rhs), get_gpu(tmp1))
loss = -cu.cusum(get_gpu(tmp1)) / N
ret = cls._create_node(loss)
ret.attrs._z = z
ret.attrs._lhs = lhs
ret.attrs._rhs = rhs
return ret
def _oper_gpu(cls, lhs, rhs):
N = lhs.shape[0]
z = get_gpu(lhs).empty_like_me()
tmp1 = get_gpu(lhs).empty_like_me()
with cu.cudnn_handler() as handle:
cu.cuSoftmaxForward(handle, lhs, z, mode=1)
cu.cucross_entropy(get_gpu(z), get_gpu(rhs), get_gpu(tmp1))
loss = -cu.cusum(get_gpu(tmp1)) / N
ret = cls._create_node(loss)
ret.attrs._z = z
ret.attrs._lhs = lhs
ret.attrs._rhs = rhs
return ret
def _oper_gpu(cls, lhs, rhs):
N = len(lhs)
z = get_gpu(lhs).empty_like_me()
tmp1 = get_gpu(lhs).empty_like_me()
tmp2 = get_gpu(lhs).empty_like_me()
cu.cusigmoid(get_gpu(lhs), z)
cu.cucross_entropy(get_gpu(z), get_gpu(rhs), tmp1)
cu.cucross_entropy(get_gpu(-z + 1.), get_gpu(-rhs + 1.), tmp2)
loss = cu.cusum(-(tmp1 + tmp2)) / N
ret = cls._create_node(loss)
ret.attrs._z = z
ret.attrs._lhs = lhs
ret.attrs._rhs = rhs
return ret
def _oper_gpu(cls, lhs, rhs):
assert rhs.ndim > 1, "Input arrays must have no less than 2 dimension."
N = len(lhs)
return cu.cusum(get_gpu((get_gpu(lhs) - get_gpu(rhs))**2)) / (N * 2)