Select columns and rows

Columns

A name returns a Series, and a list of names returns a DataFrame. The double brackets are not decoration: the outer one is "select from df" and the internal is "a list of names":

import pandas as pd

pd.set_option("display.width", 170)
pd.set_option("display.max_columns", 20)
df = pd.read_csv("dados/funcionarios_15.csv")

print(type(df["salario"]).__name__, type(df[["salario"]]).__name__)
print(df[["nome", "departamento", "salario"]].head(3))

Series DataFrame
           nome  departamento  salario
0     Ana Souza    Engenharia     4500
1    Bruno Lima     Marketing  R$ 4650
2  Carla Mendes   engenharia      4800

A list with one name is still a list, and returns a DataFrame with one column. It's the most common trick in this part.

`.loc` and `.iloc`

The .loc selects by label and the .iloc, by integer position. In a table with the default index, they seem the same, but with another index the difference appears:

print(len(df.loc[0:3]), len(df.iloc[0:3]))
print(df.iloc[-1]["nome"])

4 3
Ana Souza

The df.loc[0:3] returns 4 rows (0, 1, 2, and 3), and the df.iloc[0:3] returns 3. A slice that looks the same gives a different result, and this asymmetry catches almost everyone coming from Python slices.

An index that is not 0, 1, 2

With the id_funcionario as index, the same entry falls into different rows:

por_id = df.set_index("id_funcionario")
print(por_id.loc[103, "nome"], por_id.iloc[2]["nome"])
print(por_id.loc[103:105, "nome"].tolist())

Carla Mendes Carla Mendes
['Carla Mendes', 'Diego Rocha', 'Elisa Fernandes']

The .loc[103] searches for the label 103 (Carla), and the .iloc[2] searches for the position 2, which happens to be the same person. The slice .loc[103:105] brings the three labels, including 105. With the iloc, a number like 103 would be a position that does not exist.

An index can repeat, and here it repeats: id 101 appears twice. The .loc returns all the rows with that label, and the type of the result changes from Series to DataFrame:

print(type(por_id.loc[103]).__name__, type(por_id.loc[101]).__name__)
print(por_id.loc[101].shape)
print(por_id.index.is_unique)

Series DataFrame
(2, 9)
False

A code that assumes "one id, one row" and uses .loc[id] break (or, worse, silently passes with a result of another type) as soon as there is a repeated id. Chapter 10 resolves the repetition.

Rows and columns together

The .loc accepts rows and columns in the same call, and the condition can be a mask:

print(df.loc[df["experiencia"] > 6, ["nome", "salario"]])
print(df.loc[2:4, "nome":"cidade"])

              nome  salario
6   Gabriela Nunes     5400
7   Henrique Costa  R$ 5550
14       Ana Souza     6600
              nome      departamento          cidade
2     Carla Mendes       engenharia     Porto Alegre
3      Diego Rocha  Ciência de Dados  Rio de Janeiro
4  Elisa Fernandes                RH             NaN

For a single value, at (by label) and iat (by position) are more straightforward and faster than loc and iloc:

print(df.at[0, "nome"], df.iat[0, 1])

Ana Souza Ana Souza

Selecting also creates a new table

A selection returns an independent. Changing the result does not change the original table, and this is a guarantee of Pandas 3 (the copy on write, which Chapter 27 details):

parte = df[["nome", "idade"]]
parte.loc[0, "nome"] = "ALTERADO"
print(df.at[0, "nome"])

Ana Souza