本文共 4405 字，大约阅读时间需要 14 分钟。
In this article, we’ll take a look at using the PyTorch torch.max() function.
在本文中，我们将介绍如何使用PyTorch torch.max（）函数。
As you may expect, this is a very simple function, but interestingly, it has more than you imagine.
如您所料，这是一个非常简单的功能，但有趣的是，它具有的功能超出了您的想象。
Let’s take a look at using this function, using some simple examples.
让我们使用一些简单的示例来看看如何使用此功能。
NOTE: At the time of writing, the PyTorch version used is PyTorch 1.5.0
注意 ：在撰写本文时，使用的PyTorch版本是PyTorch 1.5.0
To use PyTorch torch.max(), first import torch.
要使用PyTorch torch.max() ，请首先导入torch 。
import torch
Now, this function returns the maximum among the elements in the Tensor.
现在，此函数返回张量中元素之间的最大值。
The default behavior is to return a single element and an index, corresponding to the global maximum element.
默认行为是返回单个元素和对应于全局最大元素的索引。
max_element = torch.max(input_tensor)
Here is an example:
这是一个例子：
p = torch.randn([2, 3])print(p)max_element = torch.max(p)print(max_element)
Output
输出量
tensor([[-0.0665, 2.7976, 0.9753], [ 0.0688, -1.0376, 1.4443]])tensor(2.7976)
Indeed, this gives us the global maximum element in the Tensor!
确实，这使我们在张量中具有全局最大元素！
However, you may wish to get the maximum along a particular dimension, as a , instead of a single element.
但是，您可能希望沿特定尺寸获得最大值，即而不是单个元素。
To specify the dimension (axis – in numpy), there is another optional keyword argument, called dim
要指定尺寸（ axis –在numpy ），还有另一个可选的关键字参数，称为dim
This represents the direction that we take for the maximum.
这代表了我们追求最大的方向。
This returns a tuple, max_elements and max_indices.
这将返回一个元组max_elements和max_indices 。
max_elements -> All the maximum elements of the Tensor. max_elements >张量的所有最大元素。 max_indices -> Indices corresponding to the maximum elements. max_indices >对应于最大元素的索引。 max_elements, max_indices = torch.max(input_tensor, dim)
This will return a Tensor, which has the maximum elements along the dimension dim.
这将返回一个Tensor，其在dim维度上具有最大元素。
Let’s now look at some examples.
现在让我们看一些示例。
p = torch.randn([2, 3])print(p)# Get the maximum along dim = 0 (axis = 0)max_elements, max_idxs = torch.max(p, dim=0)print(max_elements)print(max_idxs)
Output
输出量
tensor([[-0.0665, 2.7976, 0.9753], [ 0.0688, -1.0376, 1.4443]])tensor([0.0688, 2.7976, 1.4443])tensor([1, 0, 1])
As you can see, we find the maximum along the dimension 0 (maximum along columns).
如您所见，我们在维度0上找到最大值（在列上找到最大值）。
Also, we get the indices corresponding to the elements. For example,0.0688 has the index 1 along column 0
同样，我们获得与元素相对应的索引。 例如， 0.0688在第0列的索引为1
Similarly, if you want to find the maximum along the rows, use dim=1.
同样，如果要在行中查找最大值，请使用dim=1 。
# Get the maximum along dim = 1 (axis = 1)max_elements, max_idxs = torch.max(p, dim=1)print(max_elements)print(max_idxs)
Output
输出量
tensor([2.7976, 1.4443])tensor([1, 2])
Indeed, we get the maximum elements along the row, and the corresponding index (along the row).
确实，我们获得了沿着该行的最大元素以及相应的索引（沿着该行）。
We can also use torch.max() to get the maximum values between two Tensors.
我们还可以使用torch.max()获得两个张量之间的最大值。
output_tensor = torch.max(a, b)
Here, a and b must have the same dimensions, or must be “broadcastable” Tensors.
在此， a和b必须具有相同的尺寸，或者必须是“可广播的”张量。
Here is a simple example to compare two Tensors having the same dimensions.
这是一个比较两个具有相同尺寸的张量的简单示例。
p = torch.randn([2, 3])q = torch.randn([2, 3])print("p =", p)print("q =",q)# Compare elements of p and q and get the maximummax_elements = torch.max(p, q)print(max_elements) Output
输出量
p = tensor([[-0.0665, 2.7976, 0.9753], [ 0.0688, -1.0376, 1.4443]])q = tensor([[-0.0678, 0.2042, 0.8254], [-0.1530, 0.0581, -0.3694]])tensor([[-0.0665, 2.7976, 0.9753], [ 0.0688, 0.0581, 1.4443]])
Indeed, we get the output tensor having maximum elements between p and q.
实际上，我们得到的输出张量具有在p和q之间的最大元素。
In this article, we learned about using the torch.max() function, to find out the maximum element of a Tensor.
在本文中，我们学习了如何使用torch.max（）函数来查找张量的最大元素。
We also used this function to compare two tensors and get the maximum among them.
我们还使用此函数比较两个张量，并在其中得到最大值。
For similar articles, do go through our content on our ! Stay tuned for more!
对于类似的文章，请在我们的浏览我们的内容！ 敬请期待更多！
翻译自:
转载地址：http://yqozd.baihongyu.com/