ML Lab _ Practice _ Ellen

Front

List scenarios for scaling features.

Back

1. Regularized regression: Ensures coefficients are on the same scale for penalty term application.
2. Linear models: Helps with numerical stability in estimation.
3. Nearest neighbor methods: Prevents large magnitude features from dominating.

Front

Conditions to avoid scaling data.

Back

1. Features are in the same units or normalized.
2. Model is scale-invariant (e.g., logistic regression).

Front

Steps for standard scaling in train/val/test split.

Back

1. Split data into three groups.
2. Calculate mean and standard deviation using training data.
3. Scale training features with calculated mean and standard deviation.
4. Scale validation features with training mean and standard deviation.
5. Scale test features with training mean and standard deviation.

Front

First action in cross-validation for multi-class classification.

Back

Randomize data order.

Front

Explain the concept of cross-validated R^2.

Back

Cross-validated R^2 is the average of R^2 values obtained from K-fold cross-validation

Front

Pros of 5-fold cross-validation.

Back

Less computationally expensive, more stable estimate.

Front

Label encoding

Back

Converts categorical data into numerical format by assigning each category a distinct number

Front

Hashing trick

Back

Encodes features into fixed-size vectors using hash functions

Front

When to use one-hot encoding

Back

When the number of categories is small

Front

Advantage of label encoding

Back

Doesn’t create a huge number of columns