Core Pandas Vocabulary

Series: one labeled dimension

Everything in pandas is described with a handful of words, and using them precisely saves hours later. Start with the simpler of the two objects.

A Series is a one-dimensional labeled array that can hold values of any data type: whole numbers, decimals, text, dates, or Python objects. A Series has one axis, and its axis labels are collectively called the Index. It also has one dtype, which describes the kind of values it stores.

One axis, one set of labels, one kind of value. That is the whole definition.

The single dtype is worth pausing on, because it is the main difference between a Series and a table. A Series of temperatures is a Series of decimal numbers; a Series of city names is a Series of text. You cannot give one of its entries a type of its own.

The Index is not a store of values, and the values do not carry their own labels around independently. The Series is what binds the two.

DataFrame: two labeled axes

Most analysis needs several variables together, which is what the second object is for. A DataFrame is a two-dimensional labeled data structure whose columns can have different data types. Think of it as a table, a spreadsheet, a SQL result, or a dictionary of Series.

Its first semantic axis is the Index, which labels rows. Its second is columns, which contains the column labels. Those two axes give every cell an address made of labels rather than of numbers.

That address is the point. A DataFrame separates the position of a cell from the row and column labels that identify it, so a value is "the total of order 1002" and not "the value in the second row, fourth position".

Here the Index holds the order IDs 1001 and 1002, and columns holds customer, ordered_on, and total. Each column keeps its own type — text, date, decimal — which a single-typed grid could not do.

Rows, columns, and the labels on them

Two more words carry analytical meaning, not just structural meaning. A row represents one observation across the DataFrame's columns. A column represents one variable across the rows and can be selected as a Series.

In the orders table, the row labeled 1001 is one order seen from every angle; the column total is one variable seen across every order. Reading a table in those terms tells you immediately whether it is shaped for the question you are asking.

The row Index and the column labels form the coordinate system used to identify data, and labels do not have to be numbers. Order IDs, customer codes, dates, and country names all make good labels.

A default RangeIndex is created when a Series is built from values without an explicit index. Pass five values and no labels, and pandas supplies the labels 0 through 4 — one less than the length — so that the object always has an Index to align on.

That default is a convenience, not a signal that labels are unnecessary. Replace it as soon as the data has a meaningful key.

Structural checks before any analysis

Before calculating anything, confirm that the object really has the dimensions, labels, and types you expected. Four attributes answer that, and each answers a different question.

Use shape to read the number of rows and columns. Use index to inspect row labels and columns to inspect column labels. Use dtypes to inspect the type of each DataFrame column; use dtype for the single type of a Series.

The singular and the plural are not interchangeable, and mixing them up is the most common slip in this vocabulary.

These structural checks come before analysis because they reveal whether the object has the expected dimensions, labels, and types. A calculation on a column that silently arrived as text will not fail loudly; it will produce something that looks like a number and is not.

Containers, not displays

A useful mental model is that a DataFrame contains Series, and a Series contains scalar values. The objects are labeled containers, not merely displays of arrays: nesting runs from two dimensions, to one, to a single value.

The model never runs the other way. A Series does not contain DataFrames, and an Index does not contain either of them.

The container model has an immediate practical consequence. Selecting one DataFrame column therefore yields a Series while preserving its Index.

Take total out of the orders table and you get a Series of decimals labeled 1001 and 1002 — the same row labels the table used. Nothing has to be re-attached afterwards, and the extracted variable can still be aligned against anything else that uses those order IDs.

Handing the data to NumPy

Other libraries often expect a plain array rather than a labeled pandas object. When another library needs an array, to_numpy() returns a NumPy representation of the values.

The conversion is deliberate, because it changes what the data is. A DataFrame conversion to NumPy omits row and column labels: the numbers survive, the coordinate system does not.

There is a second effect, and it is easy to miss. Because a NumPy array has one dtype for the entire array, heterogeneous DataFrame columns may be converted to a common NumPy dtype. A table of dates, text, and decimals cannot keep three types inside one array, so pandas finds a type that can hold them all.

Use the conversion deliberately: it changes the representation and removes the labels that pandas uses to express meaning and alignment. Convert at the edge of the analysis, where a library demands it — not in the middle, where the labels are still doing work.

Key takeaways

• Series means one labeled dimension with one dtype, and those labels are its Index.
• DataFrame means labeled rows and columns whose columns may have different dtypes — a table, a spreadsheet, a SQL result, or a dictionary of Series.
• Index labels rows, columns labels columns, and shape reports the two axis lengths; dtypes reports one type per column, dtype the single type of a Series.
• A row is one observation, a column is one variable, and selecting a column returns a Series that keeps the row Index.
• Converting a DataFrame to NumPy removes its row and column labels, and mixed columns may be coerced to one common type.