Data Science interview questions and answers

Repeated looks inflate the Type I error because you get many chances to cross the significance threshold by luck.

A DataFrame is a labeled two-dimensional table with rows and named columns, similar to a spreadsheet or SQL result.

loc selects by label, including index values and column names, while iloc selects by integer position.

Use pandas.read_csv with the file path, and optionally set sep, dtype, parse_dates, and na_values.

head() returns the first few rows, five by default, so you can sanity-check columns and values.

Use bracket notation such as df['col'] or attribute access df.col when the name is a valid identifier.

A Series is one-dimensional with an index and a single dtype, like one column.

NumPy provides fast n-dimensional arrays and vectorized math that pandas and scikit-learn sit on.

An ndarray is a homogeneous, fixed-type, multidimensional array stored in contiguous memory.

Lists store pointers to Python objects, so math loops pay interpreter overhead on every element.

numpy.mean computes the arithmetic mean along an axis of an array.

SELECT chooses which columns or expressions to return from one or more tables.

WHERE filters rows before aggregation, while HAVING filters groups after GROUP BY.

GROUP BY collapses rows that share the same key into one row per group so you can compute aggregates.

An INNER JOIN returns only rows whose join keys match in both tables.

COUNT(*) counts rows, including those with nulls.

ASC sorts from smallest to largest or A to Z, and DESC reverses that order.

EDA is the process of inspecting structure, quality, distributions, and relationships before modeling or decision-making.

Wrong types break aggregations: IDs stored as floats, dates as strings, or categories as integers that get averaged.

A data dictionary documents each field's meaning, grain, allowed values, and null behavior.