One-hot encoding for text in NLP - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of one-hot encoding a small text corpus

What is the output of the following code that one-hot encodes a list of words?

NLP

from sklearn.preprocessing import OneHotEncoder
import numpy as np

words = np.array([['cat'], ['dog'], ['cat'], ['bird']])
encoder = OneHotEncoder(sparse=False)
encoded = encoder.fit_transform(words)
print(encoded)

A

[[0. 1. 0.]
 [0. 0. 1.]
 [0. 1. 0.]
 [1. 0. 0.]]

B

[[1. 0. 0.]
 [0. 1. 0.]
 [1. 0. 0.]
 [0. 0. 1.]]

C

[[0. 1. 0.]
 [1. 0. 0.]
 [0. 1. 0.]
 [0. 0. 1.]]

D

[[1. 0. 0.]
 [0. 0. 1.]
 [1. 0. 0.]
 [0. 1. 0.]]

💡 Hint

Remember that OneHotEncoder assigns columns in alphabetical order of unique words.

The unique words are ['bird', 'cat', 'dog'] alphabetically. So 'bird' is [1,0,0], 'cat' is [0,1,0], 'dog' is [0,0,1]. The input words are ['cat', 'dog', 'cat', 'bird'], so the output matches option A.

🧠 Conceptual

intermediate

Understanding one-hot encoding vocabulary size

If you one-hot encode a text corpus with 10,000 unique words, what will be the size of each one-hot vector?

A A vector of length 1 with the index of the word

B 10,000 elements with multiple elements set to 1 depending on word frequency

C A vector of length equal to the number of words in the sentence

D 10,000 elements with exactly one element set to 1 and the rest 0

💡 Hint

One-hot encoding creates a vector with one position for each unique word.

One-hot encoding creates a vector with length equal to the vocabulary size. Only one position is 1, representing the word, and all others are 0.

❓ Hyperparameter

advanced

Choosing one-hot encoding parameters for text data

Which parameter of sklearn's OneHotEncoder controls whether the output is a sparse matrix or a dense array?

A sparse

B handle_unknown

C categories

D drop

💡 Hint

This parameter decides the output format to save memory or not.

The 'sparse' parameter when set to True returns a sparse matrix, which saves memory for large vocabularies. False returns a dense numpy array.

❓ Metrics

advanced

Evaluating one-hot encoded text input for a classification model

You trained a classifier on one-hot encoded text data. Which metric best measures how well the model predicts the correct class labels?

A Perplexity

B Accuracy

C Mean Squared Error

D Silhouette Score

💡 Hint

Think about classification performance metrics.

Accuracy measures the fraction of correct predictions in classification tasks. Mean Squared Error is for regression, Silhouette Score for clustering, and Perplexity for language models.

🔧 Debug

expert

Debugging one-hot encoding with unseen words during inference

You trained a OneHotEncoder on a training set and saved it. At inference, you try to transform new text containing words not seen during training. What error will sklearn's OneHotEncoder raise by default?

A KeyError: word not found in vocabulary

B TypeError: unsupported operand type(s)

C ValueError: Found unknown categories during transform

D IndexError: index out of range

💡 Hint

Check how OneHotEncoder handles unknown categories by default.

By default, OneHotEncoder raises a ValueError if transform sees categories not present during fit. To avoid this, handle_unknown='ignore' can be set.