mindspore.ops.max

mindspore.ops.max(input, axis=None, keepdims=False, *, initial=None, where=None)

返回tensor在指定轴上的最大值及其索引。

参数：

• input (Tensor) - 输入tensor。

• axis (int) - 指定计算轴。如果为 None ，计算 input 中的所有元素。默认 None 。

• keepdims (bool) - 输出tensor是否保留维度。默认 False 。

关键字参数：

• initial (scalar, 可选) - 最大值的初始值。默认 None 。

• where (Tensor[bool], 可选) - 指定计算最大值的范围，该tensor的shape须可被广播到 input 的shape上。必须指定initial值。默认 None ，表示计算全部元素。

返回：

两个tensor组成的tuple(max, max_indices)。

支持平台：

Ascend GPU CPU

样例：

>>> import mindspore
>>> input = mindspore.tensor([[9, 3, 4, 5],
...                           [5, 2, 7, 4],
...                           [8, 1, 3, 6]])
>>> # case 1: By default, compute the maximum of all elements.
>>> mindspore.ops.max(input)
(Tensor(shape=[], dtype=Int64, value= 9),
 Tensor(shape=[], dtype=Int64, value= 0))
>>>
>>> # case 2: Compute maximum along axis 1.
>>> mindspore.ops.max(input, axis=1)
(Tensor(shape=[3], dtype=Int64, value= [9, 7, 8]),
 Tensor(shape=[3], dtype=Int64, value= [0, 2, 0]))
>>>
>>> # case 3: If keepdims=True, the output shape will be same of that of the input.
>>> mindspore.ops.max(input, axis=1, keepdims=True)
(Tensor(shape=[3, 1], dtype=Int64, value=
 [[9],
  [7],
  [8]]),
 Tensor(shape=[3, 1], dtype=Int64, value=
 [[0],
  [2],
  [0]]))
>>>
>>> # case 4: Use "where" to include only specific elements in computing the maximum.
>>> where = mindspore.tensor([[0, 0, 1, 0],
...                           [0, 0, 1, 1],
...                           [1, 1, 1, 0]], dtype=mindspore.bool_)
>>> mindspore.ops.max(input, axis=1, keepdims=True, initial=0, where=where)
(Tensor(shape=[3, 1], dtype=Int64, value=
 [[4],
  [7],
  [8]]),
 Tensor(shape=[3, 1], dtype=Int64, value=
 [[2],
  [2],
  [0]]))
>>>
>>> # case 5: The shape of "where" must be broadcast compatible with input.
>>> where = mindspore.tensor([[False],
...                           [False],
...                           [False]])
>>> mindspore.ops.max(input, axis=0, keepdims=True, initial=0, where=where)
(Tensor(shape=[1, 4], dtype=Int64, value=
 [[0, 0, 0, 0]]),
 Tensor(shape=[1, 4], dtype=Int64, value=
 [[0, 0, 0, 0]]))