Selecting with loc and iloc
"What was the branch on sale 4?" "Show rows 2 to 4, just branch and copies." A DataFrame has two ways to point at cells: by the labels printed on it, or by counting positions. pandas gives each its own tool, and mixing them up is the most common beginner bug in pandas.
The rule
locuses labels and includes the end;ilocuses positions and excludes it.
We keep using sales, now with its revenue column (8 rows; columns date, branch, genre, copies, price, revenue).
sales.loc[4, "branch"] # 'Delhi' row label 4, column named branch
sales.loc[2:4, ["branch", "copies"]] # rows labelled 2, 3 AND 4
sales.iloc[4, 1] # 'Delhi' 5th row, 2nd column
sales.iloc[2:4, [1, 3]] # rows at positions 2 and 3 onlyBoth take [rows, columns], and both accept a single item, a list, or a slice in each part. A colon alone means "all": sales.loc[:, "price"] is the whole price column.
So far labels and positions look identical, because the default index is 0, 1, 2… The difference becomes obvious once the index is something else:
by_date = sales.set_index("date")
by_date.loc["2024-03-02"] # both rows with that date label
by_date.iloc[0] # the first row, whatever its labelloc also takes a boolean mask for the rows, which is how the next lesson filters data:
sales.loc[sales["copies"] > 4, "branch"] # Delhi, ChennaiThe shortcut
Say the letter out loud: i for integer position. If you are counting ("the third row"), reach for iloc. If you are naming ("the row labelled 3", "the column called price"), reach for loc. Then apply the end rule: names include the end, counts do not.
Naming or counting?
The request says "rows 2 to 4" meaning the printed labels, and names the columns. Naming, so
loc.
The trap: sales.iloc[2:4] looks like the same request but returns only two rows, because positions stop before the end.
Drills
Answer fast — trust the rule.
How many rows does sales.loc[1:3] return?
How many rows does sales.iloc[1:3] return?
Which expression returns 'science' from row 2?
What does sales.iloc[-1, 3] return?
After by_date = sales.set_index("date"), what does by_date.loc["2024-03-03", "copies"] return?
Remember
locuses labels and includes the end;ilocuses positions and excludes it.- i for integer: counting means
iloc, naming meansloc. - An index label can repeat, so
loccan return several rows for one label.