Series: the labeled column

Create a Series

A Series can arise from a list, a dictionary, or a NumPy array. Without an explicit index, the labels are 0, 1, 2... With an index, the labels are whatever you want, and the keys of a dictionary become the index by themselves:

import numpy as np
import pandas as pd

salarios = pd.Series([4500, 6500, 5000])
print(salarios)

com_nome = pd.Series([4500, 6500, 5000], index=["Ana", "Bruno", "Carla"], name="salario")
print(com_nome)
print(pd.Series({"Ana": 4500, "Bruno": 6500}))

0    4500
1    6500
2    5000
dtype: int64
Ana      4500
Bruno    6500
Carla    5000
Name: salario, dtype: int64
Ana      4500
Bruno    6500
dtype: int64

A Series has values (the data), index (the labels) and, optionally, a name, which becomes the name of the column when it enters a DataFrame:

print(com_nome.name, com_nome.index.tolist(), com_nome.to_numpy())

salario ['Ana', 'Bruno', 'Carla'] [4500 6500 5000]

By label or by position

The index provides two ways to access a value: by label (.loc) and by position (.iloc). The two only coincide when the index is 0, 1, 2... and one should never rely on that:

print(com_nome.loc["Bruno"], com_nome.iloc[1])
print(com_nome.iloc[-1], com_nome.loc["Ana":"Bruno"].tolist())

6500 6500
5000 [4500, 6500]

Note that the slice by label ("Ana":"Bruno") includes both extremes, unlike Python slices. In Pandas 3, accessing com_nome[1] with an integer in a text index no longer falls to the position, as it did in older versions:

try:
    com_nome[1]
except KeyError as erro:
    print("KeyError:", erro)

KeyError: 1

Always use .loc or .iloc, which indicate what you want.

Vectorized operations, and alignment by label

The operations apply to the Series entire, as in NumPy. What changes is that the sum of two Series aligns by labels, and labels that only exist on one side become NaN:

print((com_nome * 1.1).round(0).tolist())
print((com_nome > 4800).tolist())
print(com_nome.mean())

bonus = pd.Series({"Ana": 500, "Carla": 300, "Diego": 100})
print(com_nome + bonus)
print(com_nome.add(bonus, fill_value=0))

[4950.0, 7150.0, 5500.0]
[False, True, True]
5333.333333333333
Ana      5000.0
Bruno       NaN
Carla    5300.0
Diego       NaN
dtype: float64
Ana      5000.0
Bruno    6500.0
Carla    5300.0
Diego     100.0
dtype: float64

Bruno has no bonus, and Diego has no salary, so the common sum gives NaN on both. The method add with fill_value=0 treats the missing side as zero, and then no one sums. This is the difference that catches most people coming from NumPy: an array sums position with position, and the Series sums label with label.

Text in a Series

The accessor .str applies text operations to all the values at once, without a loop, and let the missing ones pass:

nomes = pd.Series(["  ana souza ", "BRUNO LIMA", None])
print(nomes.str.strip().str.title().tolist())

['Ana Souza', 'Bruno Lima', nan]

An accessor is a set of tools

The .str for text, the .dt for dates (chapter 18) and the .cat for categories (chapter 19). Each one only works on the right type of column, and that is what keeps the API organized.