Deep Learning Academy · 课时

重塑、视图、压缩与扩展维度

改变维度而不复制数据

重塑、视图、压缩与扩展维度 是 CoddyKit 上的免费 Deep Learning Academy 课时。这是第 2 节课，共 4 节。你可以在下方免费阅读本课时的完整内容 — 然后在浏览器中使用内置代码编辑器和全天候 AI 导师进行实践。这是 Deep Learning Academy 学习路径的一部分，你的进度在网页和 CoddyKit 应用中同步。 Deep Learning Academy 课程共包含 4 节课。

本课时的部分内容尚未翻译，以英文显示。

Same Data, Different Shape

Often the numbers are right but the layout is wrong. Reshaping rearranges a tensor's dimensions without changing the values inside.

reshape Picks a New Layout

Call reshape with the dimensions you want. The total number of elements must stay the same.

x = torch.arange(6)
y = x.reshape(2, 3)
print(y.shape)  # torch.Size([2, 3])

Let -1 Infer a Dimension

Pass -1 for one dimension and PyTorch computes it for you from the total count. Handy when you only know the rest.

x = torch.arange(6)
y = x.reshape(-1, 2)
print(y.shape)  # torch.Size([3, 2])

view Shares the Same Memory

view reshapes without copying, so it is fast. It needs the data to be laid out contiguously in memory.

x = torch.arange(6)
y = x.view(3, 2)
print(y.shape)  # torch.Size([3, 2])

reshape Is the Safer Default

When in doubt, reach for reshape. It works even on non-contiguous tensors by copying only if it must.

Squeeze Removes Size-1 Dims

squeeze strips out any dimension of length 1. It cleans up shapes like (1, 5) down to a simple (5).

x = torch.zeros(1, 5)
y = x.squeeze()
print(y.shape)  # torch.Size([5])

Squeeze a Specific Dimension

Give squeeze an index to remove only that dimension. Safer when other size-1 dims should stay put.

x = torch.zeros(1, 5, 1)
y = x.squeeze(0)
print(y.shape)  # torch.Size([5, 1])

Unsqueeze Adds a Dimension

unsqueeze inserts a new size-1 dimension at the position you choose. It is the exact opposite of squeeze.

x = torch.tensor([1, 2, 3])
y = x.unsqueeze(0)
print(y.shape)  # torch.Size([1, 3])

Why You Add a Batch Dimension

Models expect a batch dimension up front. unsqueeze(0) turns one sample into a batch of one so the model accepts it.

sample = torch.randn(3)
batch = sample.unsqueeze(0)
print(batch.shape)  # torch.Size([1, 3])

Flatten Down to One Line

flatten collapses every dimension into a single long vector. It is common right before a final linear layer.

x = torch.zeros(2, 3)
y = x.flatten()
print(y.shape)  # torch.Size([6])

Element Count Never Changes

Every reshape trick keeps the same total number of values. If the counts don't match, PyTorch raises a shape error.

Quick Check

Let's see if you can predict the shape changes.

Recap: Reshape, View, Squeeze & Unsqueeze

You can now bend tensors into any layout: reshape for flexibility, view for speed, and squeeze or unsqueeze to drop and add dimensions. 🔧

常见问题解答

「重塑、视图、压缩与扩展维度」课时是免费的吗？

是的 — 「重塑、视图、压缩与扩展维度」的完整文本可在网页上免费阅读。要进行交互式练习（内置代码编辑器和全天候 AI 导师）并解锁 Deep Learning Academy 课程的其余内容，请升级到 CoddyKit PRO。 Deep Learning Academy 课程共包含 4 节课。

「重塑、视图、压缩与扩展维度」这节课中我会学到什么？

改变维度而不复制数据你通过在浏览器中直接运行的动手代码来练习 Deep Learning Academy，全天候 AI 导师会在你学习这节课的过程中回答你的问题。

学习 Deep Learning Academy 需要有经验吗？

无需任何先前经验。CoddyKit 上的 Deep Learning Academy 课程适合初学者到高级学习者，你可以从这里开始或从头开始，按照自己的节奏学习。这是第 2 节课，共 4 节。

「重塑、视图、压缩与扩展维度」课时需要多长时间？

大多数 CoddyKit 课程大约需要 5–10 分钟。每节课都很精短且互动，所以你能稳步进步，并在网页和应用中从离开的地方继续。

我能在这节 Deep Learning Academy 课中编写并运行代码吗？

能。每节 Deep Learning Academy 课都包含内置代码编辑器，你可以在浏览器中直接编写并运行真实代码，并获得即时 AI 反馈 — 无需本地设置。

此课程中的所有课时

← 返回 Deep Learning Academy