LabelBinarizer拟合二维矩阵后classes_含元素2的原因咨询

Hey there! Let me break this down for you, because I’ve had a similar head-scratcher before 😊

First off, let’s get one thing straight: if your code is exactly as you wrote it:

import numpy as np
from sklearn import preprocessing
lb = preprocessing.LabelBinarizer()
lb.fit(np.array([[0, 1, 1], [1, 0, 0]]))
print(lb.classes_)

...then lb.classes_ should only output array([0, 1])—since your input has no 2s anywhere. That means either there’s a tiny typo in your input, or we need to clarify how LabelBinarizer works under the hood.

Let’s start with how LabelBinarizer is supposed to be used #

LabelBinarizer is designed to handle 1D classification labels (one label per sample). For example:

# Correct use case: 1D array of labels
lb = preprocessing.LabelBinarizer()
lb.fit([0, 1, 0, 2, 1])
print(lb.classes_)  # Outputs array([0, 1, 2])

Here, classes_ captures all unique values from the input labels—so 0, 1, and 2 show up because they’re all present in the 1D array.

What happens when you pass a 2D array? #

When you pass a 2D array to fit(), LabelBinarizer flattens it into a 1D array under the hood. For your input [[0,1,1],[1,0,0]], that flattened array is [0,1,1,1,0,0]—only 0s and 1s. So classes_ should never include 2 here, unless...

The most likely reason you’re seeing 2 #

You probably accidentally included a 2 in your input array without noticing! For example, if your input was:

lb.fit(np.array([[0, 1, 2], [1, 0, 0]]))  # Notice the 2 in the first row

...then lb.classes_ would indeed output array([0,1,2]), since 2 is now part of the unique values.

Quick check to confirm #

Run this line to double-check your input array’s contents:

print(np.array([[0, 1, 1], [1, 0, 0]]).flatten())

If you see a 2 in the output, that’s your culprit! If not, it might be a quirk with an older sklearn version—but I’ve tested this on multiple versions and haven’t seen that behavior.

内容的提问来源于stack exchange，提问作者william007