James Powell - How Dimensional is a `pandas.DataFrame`, anyway? | PyData Amsterdam 2024

A Pandas DataFrame is fundamentally one-dimensional data with a hierarchical index, despite often being described as two-dimensional

The distinction between structural coordinates (fixed, countable, human-scale) and data coordinates (variable, uncountable, automatable) is key to understanding DataFrame dimensionality

Index alignment is a core feature of Pandas - it’s about operating on one-dimensional, index-aligned collections of data

Group by operations and stack/unstack are essentially equivalent - they’re both about turning one homogeneous dataset into multiple datasets

Lists in Python are fundamentally one-dimensional and loosely homogeneous, while tuples represent one thing with multiple aspects

NumPy arrays are fixed-size and strictly homogeneous, providing an interpretive view of contiguous memory

Pandas operations are optimized for working down the index, not across columns - this affects performance and API design

Prices and financial data are inherently non-linear - proper modeling requires understanding this limitation

Multi-leg trades and complex financial operations are better modeled as one-dimensional series with appropriate indexing than forced into two dimensions

The “two-dimensional” nature of DataFrames is more about convenience in representation than actual dimensionality of the underlying data structure