Python 2

Open Ended

Why is it important to avoid data leakage when applying imputation strategies in data cleaning?

Multiple Choice

Which of the following is a possible effect of mean or median imputation on data distribution?

• It increases the variance of the data

• It shrinks the variance of the data

• It creates more missing values

• It has no effect on the data

Multiple Choice

Which imputation method is generally preferred when the data distribution is highly skewed, such as income?

• Mean imputation

• Median imputation

• Mode imputation

• Random imputation

Open Ended

Describe a scenario where using a special category like 'Unknown' for categorical imputation might be more informative than using the mode.

Multiple Select

Which of the following are objectives when handling duplicate records in a dataset?

• Distinguish entity vs event duplicates

• Design deduplication rules with business logic

• Aggregate events to entity-level features

• Normalize schema types

Fill in the Blank

Fill in the blank: Multiple logs of the same event, such as retries or re-sends, are called ___ duplicates.

Open Ended

What is the difference between full-row and subset-based duplicate detection, and when would you use each approach?

Multiple Select

Which of the following are important steps to ensure data type consistency in a pandas DataFrame?

• Convert numeric strings and currencies safely

• Handle mixed-type columns

• Parse datetimes and extract features

• Aggregate duplicate events

Multiple Choice

Which of the following is NOT a problem caused by incorrect data types in machine learning pipelines?

• Aggregations on string numerics concatenate instead of summing

• Mixed-type columns become 'object', hiding structure

• Broken datetime parsing leads to NaT and missing features

• All numeric columns are automatically converted to float

Open Ended

Explain how using pd.to_numeric(errors='coerce') helps in handling messy numeric data entries. What are the benefits and what should you inspect after conversion?

Multiple Select

Which steps are recommended for cleaning currency data before converting to numeric types? Select all that apply.

• Remove non-numeric characters except dot and minus

• Convert with to_numeric(errors='coerce')

• Treat failed parses as missing

• Ignore validation of result ranges

Fill in the Blank

Single datetime64 dtype enables time features, invalid entries become ___, and format can be inferred automatically.

Open Ended

Why is it important for feature-generation functions to be identical in both offline training and online inference when extracting features from dates?

Multiple Choice

Which method can be used to mitigate the impact of extreme outliers in a dataset?

• IQR and z-score methods

• np.log1p transformation

• Clipping values

• All of the above

Multiple Choice

Which of the following best describes the difference between statistical outliers and domain outliers?

• Statistical outliers are always errors, while domain outliers are always valid.

• Statistical outliers are values far from the bulk of the distribution, while domain outliers are determined by domain knowledge.

• Domain outliers are detected using IQR, while statistical outliers are detected using Z-score.

• Statistical outliers are always valid, while domain outliers are always errors.

Multiple Select

Which of the following are true about the IQR (Interquartile Range) method for outlier detection?

• It is based on quartiles and not affected by extreme values.

• It uses the mean and standard deviation to detect outliers.

• It is the standard method for box plots.

• It is less effective for symmetric distributions.

Fill in the Blank

Fill in the blank: The Z-score method for outlier detection assumes approximate ___ in the data.

Multiple Choice

After handling outliers using removal, capping, or transformation, what is a potential risk when removing outliers that are actually valid extreme cases?

• The dataset becomes more balanced.

• Important signals may be lost.

• The mean and median become equal.

• The model becomes less sensitive to rare events.

Open Ended

Describe a scenario where capping could disadvantage a specific group and how you would detect that risk.

Open Ended

Explain why it is important to inspect missingness, duplicates, and outliers by subgroup rather than only globally when cleaning data.

Multiple Choice

Which of the following is NOT a key pattern to check during data cleaning as per the checklist?

• Missing Values

• Imputation

• Model Training

• Duplicates

Fill in the Blank

Outliers may represent specific subgroups, leading to potential ___ in data analysis.

Multiple Choice

Which of the following strategies is used for handling outliers in real datasets?

• Capping and flooring

• Ignoring outliers

• Random sampling

• Feature scaling

Multiple Choice

When is it reasonable to use capping for outliers instead of removing them?

• When tails are real but shouldn't dominate

• When the goal is robust prediction

• When sample size is limited

• All of the above

Multiple Choice

Which of the following is a potential effect of mean or median imputation on data distribution?

• It increases variance

• It shrinks variance

• It creates artificial spikes

• It strengthens all correlations

Open Ended

Reflecting on today's lesson about imputation strategies, what is one key takeaway you have about why imputation is never neutral?