max

paddle. max ( x: Tensor, axis: int | Sequence[int] | None = None, keepdim: bool = False, name: str | None = None ) Tensor [source]

Computes the maximum of tensor elements over the given axis.

Note

The difference between max and amax is: If there are multiple maximum elements, amax evenly distributes gradient between these equal values, while max propagates gradient to all of them.

Parameters :
  • x (Tensor) – A tensor, the data type is float8_e4m3fn, float8_e5m2, bfloat16, float16, float32, float64, int32, int64.

  • axis (int|list|tuple|None, optional) – The axis along which the maximum is computed. If None, compute the maximum over all elements of x and return a Tensor with a single element, otherwise must be in the range \([-x.ndim(x), x.ndim(x))\). If \(axis[i] < 0\), the axis to reduce is \(x.ndim + axis[i]\).

  • keepdim (bool, optional) – Whether to reserve the reduced dimension in the output Tensor. The result tensor will have one fewer dimension than the x unless keepdim is true, default value is False.

  • name (str|None, optional) – Name for the operation (optional, default is None). For more information, please refer to Name.

Returns :

Tensor, results of maximum on the specified axis of input tensor, it’s data type is the same as x.

Examples

>>> import paddle

>>> # data_x is a Tensor with shape [2, 4]
>>> # the axis is a int element
>>> x = paddle.to_tensor([[0.2, 0.3, 0.5, 0.9],
...                       [0.1, 0.2, 0.6, 0.7]],
...                       dtype='float64', stop_gradient=False)
>>> result1 = paddle.max(x)
>>> result1.backward()
>>> result1
Tensor(shape=[], dtype=float64, place=Place(cpu), stop_gradient=False,
0.90000000)
>>> x.grad
Tensor(shape=[2, 4], dtype=float64, place=Place(cpu), stop_gradient=False,
[[0., 0., 0., 1.],
 [0., 0., 0., 0.]])

>>> x.clear_grad()
>>> result2 = paddle.max(x, axis=0)
>>> result2.backward()
>>> result2
Tensor(shape=[4], dtype=float64, place=Place(cpu), stop_gradient=False,
[0.20000000, 0.30000000, 0.60000000, 0.90000000])
>>> x.grad
Tensor(shape=[2, 4], dtype=float64, place=Place(cpu), stop_gradient=False,
[[1., 1., 0., 1.],
 [0., 0., 1., 0.]])

>>> x.clear_grad()
>>> result3 = paddle.max(x, axis=-1)
>>> result3.backward()
>>> result3
Tensor(shape=[2], dtype=float64, place=Place(cpu), stop_gradient=False,
[0.90000000, 0.70000000])
>>> x.grad
Tensor(shape=[2, 4], dtype=float64, place=Place(cpu), stop_gradient=False,
[[0., 0., 0., 1.],
 [0., 0., 0., 1.]])

>>> x.clear_grad()
>>> result4 = paddle.max(x, axis=1, keepdim=True)
>>> result4.backward()
>>> result4
Tensor(shape=[2, 1], dtype=float64, place=Place(cpu), stop_gradient=False,
[[0.90000000],
 [0.70000000]])
>>> x.grad
Tensor(shape=[2, 4], dtype=float64, place=Place(cpu), stop_gradient=False,
[[0., 0., 0., 1.],
 [0., 0., 0., 1.]])

>>> # data_y is a Tensor with shape [2, 2, 2]
>>> # the axis is list
>>> y = paddle.to_tensor([[[1.0, 2.0], [3.0, 4.0]],
...                         [[5.0, 6.0], [7.0, 8.0]]],
...                         dtype='float64', stop_gradient=False)
>>> result5 = paddle.max(y, axis=[1, 2])
>>> result5.backward()
>>> result5
Tensor(shape=[2], dtype=float64, place=Place(cpu), stop_gradient=False,
[4., 8.])
>>> y.grad
Tensor(shape=[2, 2, 2], dtype=float64, place=Place(cpu), stop_gradient=False,
[[[0., 0.],
  [0., 1.]],
 [[0., 0.],
  [0., 1.]]])

>>> y.clear_grad()
>>> result6 = paddle.max(y, axis=[0, 1])
>>> result6.backward()
>>> result6
Tensor(shape=[2], dtype=float64, place=Place(cpu), stop_gradient=False,
[7., 8.])
>>> y.grad
Tensor(shape=[2, 2, 2], dtype=float64, place=Place(cpu), stop_gradient=False,
[[[0., 0.],
  [0., 0.]],
 [[0., 0.],
  [1., 1.]]])