Data Science Academy · 课时

先缩放，再拟合 PCA

为什么标准化在这里很重要

先缩放，再拟合 PCA 是 CoddyKit 上的免费 Data Science 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 Data Science Academy 学习路径的一部分，你的进度在网页和 CoddyKit 应用中同步。 Data Science 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。

PCA Cares About Scale

PCA chases the largest variance, but variance depends on units. A feature in big numbers can dominate just because of its scale.

A Unit Trap

Imagine salary in dollars beside age in years. Salary varies in thousands, so PCA would treat it as far more important, unfairly.

Standardize First

The fix is standardization: rescale every feature to mean zero and unit variance so each starts on equal footing.

Use StandardScaler

In scikit-learn, StandardScaler centers and scales your columns before PCA ever sees them.

from sklearn.preprocessing import StandardScaler
Xs = StandardScaler().fit_transform(X)

Then Fit PCA

Run PCA on the scaled data, never the raw data. Now each component reflects real structure, not lopsided units.

from sklearn.decomposition import PCA
scores = PCA(n_components=2).fit_transform(Xs)

Centering Matters Too

PCA assumes data is centered at zero. Standardizing handles this by subtracting each column mean, so axes pass through the data center.

Chain It in a Pipeline

Wrap the scaler and PCA in a Pipeline so scaling always happens first and never leaks across your data splits.

from sklearn.pipeline import make_pipeline
pipe = make_pipeline(StandardScaler(), PCA(2))

Fit on Train Only

Fit the scaler on training data, then reuse it on test data. Fitting on everything leaks information and inflates your scores.

When You May Skip It

If every feature already shares the same unit and range, scaling matters less. When unsure, standardize anyway; it rarely hurts.

Watch the Difference

Run PCA with and without scaling on mixed-unit data. The explained variance and top components will look strikingly different.

Robust Options Exist

With strong outliers, consider RobustScaler, which uses medians and quartiles so extreme points sway PCA less.

from sklearn.preprocessing import RobustScaler

Quick Check

One step almost always comes right before PCA.

Recap

PCA is scale-sensitive, so standardize first, fit on training data only, and a pipeline keeps the order safe. 🎯

常见问题解答

「先缩放，再拟合 PCA」课时是免费的吗？

是的 — 「先缩放，再拟合 PCA」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 Data Science Academy 课程的其余内容，请升级到 CoddyKit PRO。 Data Science Academy 课程共包含 4 节课。

「先缩放，再拟合 PCA」这节课中我会学到什么？

为什么标准化在这里很重要你通过在浏览器中直接运行的动手代码来练习 Data Science 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 Data Science Academy 需要有经验吗？

无需任何先前经验。CoddyKit 上的 Data Science 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。

「先缩放，再拟合 PCA」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 Data Science Academy 课中编写并运行代码吗？

能。每节 Data Science 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 Data Science Academy