CatBoost in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is the main advantage of CatBoost's handling of categorical features?

CatBoost is known for its special way of dealing with categorical data. What is the main advantage of this approach compared to traditional methods like one-hot encoding?

A It ignores categorical features to speed up training.

B It converts categories into numerical values using target statistics without data leakage.

C It converts categories into random numbers to increase model randomness.

D It requires manual one-hot encoding before training.

💡 Hint

Think about how CatBoost prevents information from leaking from the target variable into the feature transformation.

CatBoost uses a technique called ordered target statistics to convert categorical features into numbers. This method avoids data leakage by calculating statistics in a way that does not use the target value of the current row during training.

❓ Model Choice

intermediate

Choosing CatBoost for a dataset with many categorical features

You have a dataset with 50 categorical features and 10 numerical features. You want a model that handles categorical data well without extensive preprocessing. Which model is the best choice?

A Linear regression, because it is simple and fast.

B Support Vector Machine, because it works well with categorical data.

C K-Nearest Neighbors, because it handles categorical data well by default.

D CatBoost, because it natively supports categorical features.

💡 Hint

Consider which model can directly use categorical features without manual encoding.

CatBoost is designed to handle categorical features natively, which means it can use them directly without manual encoding. Other models usually require preprocessing steps like one-hot encoding.

❓ Metrics

advanced

Interpreting CatBoost training output metrics

During CatBoost training, you see the following output for a binary classification task:

Iteration 100: train Logloss = 0.25, validation Logloss = 0.30

What does this tell you about the model's performance?

A The model perfectly fits both training and validation data.

B The model is underfitting because both losses are low.

C The model fits training data well but may be overfitting since validation loss is higher.

D The model is not learning because validation loss is higher than training loss.

💡 Hint

Compare training and validation losses to understand model fit.

The training loss is lower than validation loss, which suggests the model fits training data well but may not generalize perfectly to unseen data, indicating possible overfitting.

🔧 Debug

advanced

Why does CatBoost training raise a 'CatBoostError: Categorical feature is not converted'?

You try to train a CatBoost model with categorical features but get this error:

CatBoostError: Categorical feature is not converted

What is the most likely cause?

A You did not specify which columns are categorical in the CatBoost Pool or model parameters.

B Your dataset contains missing values in numerical columns.

C You used a regression loss function for classification.

D You did not normalize the numerical features before training.

💡 Hint

Think about how CatBoost knows which features are categorical.

CatBoost requires you to explicitly tell it which features are categorical by passing their indices or names. If you don't, it treats them as numerical and raises this error when it encounters unexpected data types.

❓ Hyperparameter

expert

Optimizing CatBoost with the 'depth' hyperparameter

You want to improve your CatBoost model's performance on a complex dataset. You consider increasing the 'depth' parameter from 6 to 10. What is the most likely effect?

A The model may capture more complex patterns but risks overfitting and longer training time.

B The model will train faster and generalize better with higher depth.

C The model's performance will not change because depth only affects numerical features.

D The model will ignore categorical features with higher depth.

💡 Hint

Think about how tree depth affects model complexity and training time.

Increasing tree depth allows the model to learn more complex patterns but can lead to overfitting and slower training because trees become larger and more detailed.