Target encoding in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why use target encoding instead of one-hot encoding?

Which of the following is the main advantage of using target encoding for categorical variables compared to one-hot encoding?

A Target encoding always prevents overfitting better than one-hot encoding.

B Target encoding converts categorical variables into multiple binary columns.

C Target encoding reduces dimensionality by converting categories into a single numeric value based on the target variable.

D Target encoding does not require the target variable during training.

💡 Hint

Think about how target encoding transforms categories compared to one-hot encoding.

Target encoding replaces each category with a number derived from the target variable, reducing the number of features. One-hot encoding creates many binary columns, increasing dimensionality.

❓ Predict Output

intermediate

Output of target encoding with smoothing

Given the following data and target encoding with smoothing, what is the encoded value for category 'B'?

ML Python

import pandas as pd

data = pd.DataFrame({'category': ['A', 'B', 'B', 'C', 'A', 'B'], 'target': [1, 0, 1, 0, 1, 1]})

# Global mean of target
global_mean = data['target'].mean()

# Calculate category mean and count
category_stats = data.groupby('category')['target'].agg(['mean', 'count'])

# Smoothing factor
alpha = 2

# Calculate smoothed mean
category_stats['smoothed'] = (category_stats['mean'] * category_stats['count'] + global_mean * alpha) / (category_stats['count'] + alpha)

encoded_value_B = category_stats.loc['B', 'smoothed']
print(round(encoded_value_B, 3))

A 0.5

B 0.667

C 0.75

D 0.4

💡 Hint

Calculate the category mean and count for 'B', then apply the smoothing formula.

Category 'B' has target values [0,1,1], mean=0.667, count=3. Global mean=0.667. Smoothing: (0.667*3 + 0.667*2)/(3+2) = (2 + 1.333)/5 = 0.667. Rounded to 0.667.

❓ Model Choice

advanced

Choosing target encoding for model type

For which type of model is target encoding most beneficial compared to one-hot encoding?

A Neural networks that require one-hot encoded inputs.

B Tree-based models like random forests that handle categorical variables natively.

C K-nearest neighbors models that rely on distance metrics.

D Linear regression models that assume numeric input features.

💡 Hint

Consider how different models interpret numeric vs categorical features.

Linear models benefit from target encoding because it converts categories into meaningful numeric values. Tree models can handle categories without encoding, and neural networks often prefer one-hot encoding.

❓ Metrics

advanced

Effect of target encoding on model evaluation metrics

What is a common risk when using target encoding without proper cross-validation, and how does it affect evaluation metrics?

A Data leakage causing overly optimistic metrics like accuracy or RMSE on validation data.

B Underfitting leading to poor training accuracy but good validation accuracy.

C Increased model bias causing metrics to be consistently low on both train and test sets.

D No effect on metrics because target encoding is independent of the target variable.

💡 Hint

Think about what happens if the target information leaks into the features during training.

Target encoding uses the target variable to encode features. If encoding is done on the whole dataset before splitting, it leaks target info into features, causing inflated evaluation metrics.

🔧 Debug

expert

Debugging target encoding implementation error

Consider this code snippet for target encoding. What error will it raise when run?

ML Python

import pandas as pd

data = pd.DataFrame({'cat': ['x', 'y', 'x', 'z'], 'target': [1, 0, 1, 0]})

means = data.groupby('cat')['target'].mean()

# Incorrect: trying to map means to original data without converting to dict
encoded = data['cat'].map(means)

print(encoded)

A No error; prints encoded values correctly.

B KeyError because 'cat' values are missing in means index.

C TypeError because map expects a dict but gets a Series.

D ValueError due to mismatched index during mapping.

💡 Hint

Check if pandas Series can be used directly with map for mapping values.

pandas.Series can be used directly with map to replace values based on index. No error occurs and encoded prints correctly.