Deep Learning Academy · 课时

帮您省去循环的广播规则

逐元素组合形状不同的张量

帮您省去循环的广播规则 是 CoddyKit 上的免费 Deep Learning Academy 课时。这是第 3 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。

Combine Without Matching Shapes

Broadcasting lets PyTorch stretch a smaller tensor to fit a bigger one, so you skip writing loops to repeat values.

Add a Scalar to Everything

The simplest broadcast: add one number to a whole tensor. PyTorch applies it to every element at once. ✨

x = torch.tensor([1, 2, 3])
print(x + 10)  # tensor([11, 12, 13])

Compare Shapes Right to Left

Broadcasting lines up dimensions from the right. It then checks each pair to decide if they can combine.

Rule One: Equal Sizes Match

Two dimensions are compatible when they are equal. Matching sizes line up one to one with no stretching needed.

Rule Two: A Size of 1 Stretches

If one dimension is 1, it expands to match the other. That single value is reused across the whole axis.

a = torch.tensor([[1], [2], [3]])
b = torch.tensor([10, 20])
print((a + b).shape)  # torch.Size([3, 2])

Missing Dimensions Count as 1

When one tensor has fewer dimensions, PyTorch pads it with leading ones. A vector can broadcast against a matrix this way.

m = torch.ones(2, 3)
v = torch.tensor([1, 2, 3])
print((m + v).shape)  # torch.Size([2, 3])

Add a Bias Across Rows

A classic use: add a row bias to every sample in a batch. One small vector reaches every row for free.

batch = torch.zeros(4, 3)
bias = torch.tensor([1.0, 2.0, 3.0])
print((batch + bias).shape)  # torch.Size([4, 3])

Normalize a Column at Once

Subtract a per-column mean with broadcasting and center your data in one line, no loop over rows required.

x = torch.tensor([[1.0, 2.0], [3.0, 4.0]])
m = x.mean(dim=0)
print(x - m)

When Shapes Don't Broadcast

If two dimensions differ and neither is 1, the broadcast fails and PyTorch raises a clear size error.

Add Dimensions to Steer Broadcasting

Use unsqueeze to insert a size-1 axis exactly where you need it. This guides broadcasting toward the shape you want.

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

Why Broadcasting Is Fast

Broadcasting never copies the stretched values, it reuses them. That makes vectorized math far faster than Python loops. 🚀

Quick Check

Can you predict whether two shapes will broadcast?

Recap: Broadcasting Rules

You learned the two rules: dimensions match when equal or when one is 1. Master this and loops melt into clean, fast tensor math. 💪

常见问题解答

「帮您省去循环的广播规则」课时是免费的吗？

是的 — 「帮您省去循环的广播规则」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。

「帮您省去循环的广播规则」这节课中我会学到什么？

逐元素组合形状不同的张量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 3 节课，共 4 节。

「帮您省去循环的广播规则」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