Microsoft Technical Round - MCQ Round

What is the purpose of Universal Functions (ufuncs) in NumPy?

• They are functions that can be applied universally to any Python object.

• They are vectorized wrappers around C functions that perform element-wise operations on NumPy arrays.

• They provide a generic interface for integrating with other scientific computing libraries.

• They are functions designed for processing string data within NumPy arrays.

What is the fundamental difference in data structure between a Pandas Series and a Pandas DataFrame?

• A Series is immutable, while a DataFrame is mutable.

• A Series is a 1-dimensional labeled array, whereas a DataFrame is a 2-dimensional labeled data structure with columns of potentially different types.

• A Series can only hold numerical data, while a DataFrame can hold any data type.

• A Series is built on top of Python lists, while a DataFrame is built on top of NumPy arrays.

How does Pandas primarily handle missing data (NaN values) during most arithmetic operations or aggregations (e.g., sum(), mean())?

• It raises an error if any NaN values are present.

• It automatically replaces NaN values with 0 before calculation.

• It typically skips or excludes NaN values from the calculation by default.

• It converts the entire column to a string type if NaN values exist.

What is the advantage of using vectorized operations (e.g., df['col'] + 5) in Pandas compared to iterating row by row with a for loop?

• Vectorized operations use less memory.

• Vectorized operations are usually implemented in highly optimized C or Cython, making them significantly faster.

• for loops are explicitly discouraged in Pandas and will always raise a warning.

• Vectorized operations provide more control over the iteration process.

You are training a neural network and observe that the training loss is decreasing, but the validation loss starts increasing after a certain number of epochs.

• Increase the learning rate significantly.

• Add more layers to the neural network.

• Implement early stopping or add regularization (e.g., L1/L2, dropout).

• Increase the size of the training dataset.

When is a "Bagging" ensemble method generally more effective than a "Boosting" ensemble method?

• When the base learners are inherently strong and prone to high variance.

• When the base learners are weak and prone to high bias.

• When the dataset is very small and prone to noise.

• When the goal is to interpret the contribution of individual base learners.

Which of the following scenarios would typically benefit most from using a kernel trick (e.g., in SVMs)?

• When the data is linearly separable in its original feature space.

• When the number of features is extremely high, and you need dimensionality reduction.

• When the data is non-linearly separable in its original feature space but potentially separable in a higher-dimensional space.

• When you need a highly interpretable model.

You're building a recommendation system and decide to use Matrix Factorization (e.g., SVD). What is the primary conceptual benefit of this approach over a purely collaborative filtering method based on user-item similarity?

• It handles sparse data more effectively by learning latent features, which can capture underlying user preferences and item characteristics.

• It guarantees perfect recommendations for new users (cold start problem).

• It is computationally less expensive for extremely large datasets.

• It directly incorporates content-based features without modification.

A data analyst needs to calculate the average sales for each product category while retaining all individual transaction rows. Which SQL construct is inherently designed for this specific analytical need,

• A standard GROUP BY clause, because it aggregates data by category.

• A subquery that calculates AVG(sales) per category, then JOINed back to the original table, because it pre-calculates the average.

• A Common Table Expression (CTE) defining the category average, then JOINed, because CTEs improve readability.

• A window function (e.g., AVG(sales) OVER (PARTITION BY Category)), because it computes an aggregate value for a group of rows while each row retains its distinct identity.

A database is strictly normalized (e.g., 3NF). What is a common analytical challenge or trade-off introduced by high normalization when performing complex reporting queries?

• Data integrity is compromised due to redundant data storage.

• Queries for aggregate reports or complex relationships often require numerous JOIN operations across multiple tables, potentially impacting query performance and complexity.

• It makes it impossible to store historical data.

• It prevents the use of indexes, slowing down data retrieval.