创建 MultiIndex

A MultiIndex (also called a hierarchical index) allows a Pandas DataFrame or Series to have multiple levels of row labels. Think of it as a composite key in a database: instead of identifying a row by a single label, you identify it by a tuple like (country, city) or (year, quarter). MultiIndexes are essential for representing panel data, cross-sectional time series, and any dataset with a natural two-level grouping structure.

pd.MultiIndex.from_tuples(list_of_tuples, names=) is the most explicit way to create a MultiIndex. Each tuple becomes one row label, and its elements become the levels. The names parameter assigns a label to each level (e.g. ['year', 'quarter']). This method is useful when you have a pre-built list of composite keys that you want to use as the row index.

pd.MultiIndex.from_product(iterables, names=) creates a MultiIndex from the Cartesian product of multiple lists — every combination of elements from the input lists. This is the most convenient method when all combinations of levels should exist in your data. For example, all combinations of 3 years × 4 quarters × 5 regions creates a 60-row balanced panel automatically.

The most common way to create a MultiIndex in practice is to start with a regular DataFrame that has grouping columns, then call df.set_index([col1, col2]). This promotes those columns from data columns into index levels. The result is the same as if you had built the index from scratch with from_tuples, but it takes only one line starting from tabular data.

A MultiIndex stores its data in levels (the unique values at each level) and codes (integer pointers into the level arrays). Inspect the levels with df.index.levels or df.index.get_level_values(level). Use df.index.nlevels to check how many levels exist. The names attribute lists the name assigned to each level — set these with df.index.set_names(['level0', 'level1']).

A MultiIndex can also be applied to columns, creating a hierarchy of column labels. This is common when you store multiple metrics for each category in a pivot-style layout — for example, (revenue, Q1), (revenue, Q2), (cost, Q1), (cost, Q2). Access columns with a MultiIndex the same way you access rows — using tuples. Create a column MultiIndex with pd.MultiIndex.from_product and assign it to df.columns.

When you call groupby([col1, col2]).agg(...), the resulting DataFrame has a MultiIndex on its rows (the two groupby keys become the two index levels) and potentially a MultiIndex on its columns (if you aggregate multiple metrics). This is the most common way to encounter a MultiIndex in everyday data analysis — understanding how to navigate it is essential for working with groupby results.

Use df.swaplevel() to swap two index levels, and df.reorder_levels([...]) to change the order of all levels. Reordering is useful when you want to slice by an inner level first, or when two DataFrames have their index levels in different order and need to be aligned before merging. After reordering, always call sort_index() to restore lexicographic order for efficient slicing.

Slicing into a MultiIndex is only efficient when the index is sorted lexicographically. Check whether the index is sorted with df.index.is_monotonic_increasing. If it returns False, sort with df.sort_index(). An unsorted MultiIndex triggers a PerformanceWarning on slice operations and can return incorrect results in some edge cases — always sort after creating or modifying a MultiIndex.

Call df.reset_index() to convert all index levels back into regular columns, leaving the DataFrame with a plain integer RangeIndex. This is a common pattern after groupby aggregations: compute the grouped summary with a MultiIndex, then reset the index to get a flat DataFrame suitable for further Pandas operations, merging, or export to CSV. Use reset_index(level='name') to reset only one level of a two-level index.

Use a MultiIndex when your data has a natural hierarchical structure: time series with sub-daily granularity (date + hour), panel data (entity + time period), or nested groupings (country + region + city). Avoid MultiIndexes for simple two-column group keys where a flat DataFrame with separate columns is equally readable. If you find yourself constantly calling reset_index() just to access data, that is a sign the MultiIndex is not adding value in your workflow.

Test your understanding of MultiIndex creation from this lesson.

In this lesson you learned: MultiIndexes provide hierarchical row (or column) labels using tuples, created with from_tuples, from_product, or set_index on multiple columns, and always sorted for efficient slicing. Next up we explore how to select data from a MultiIndex using .loc, tuples, slices, and the IndexSlice helper.