Selecting rows and columns with loc, iloc and filters

Most pandas screening questions about selection come down to one idea: loc works with labels and iloc works with positions. If you can predict the output of a slice before running it, you will get these questions right.

loc uses labels

loc takes row labels first and column labels second. With the default index of 0, 1, 2 and so on, labels and positions look the same, which is why people confuse the two.

df.loc[0:2, ["name", "city"]]   # rows labelled 0, 1 and 2
df.loc[df["city"] == "Lagos", "sales"]

A loc slice includes the end label. So df.loc[0:2] returns three rows. If the index holds dates or employee codes, you slice with those values, such as df.loc["2024-01-01":"2024-01-31"] on a date index.

iloc uses positions

iloc behaves like Python list slicing. The end position is excluded, and negative numbers count from the end.

df.iloc[0:2]      # first two rows
df.iloc[-1]       # last row
df.iloc[:, 0:3]   # first three columns

A common mistake is passing a column name to iloc. It only accepts integers, slices of integers or boolean arrays, so df.iloc[0, "name"] fails.

Filtering with conditions

A condition such as df["age"] > 30 returns a True/False Series, and passing it inside brackets keeps the True rows. To combine conditions:

• Use & for and, | for or, and ~ for not.
• Put each condition in its own parentheses, because & is evaluated before comparison operators.
• Use isin for several allowed values: df[df["city"].isin(["Pune", "Nairobi"])].

Writing and instead of & raises a ValueError about the truth value of a Series being ambiguous. Interviewers like this error because it shows whether you have actually written filters yourself.

Changing values safely

To update filtered rows, do it in one loc call:

df.loc[df["age"] < 0, "age"] = 0

Chained indexing such as df[df["age"] < 0]["age"] = 0 may change a temporary copy instead of df.

Habit to build: before running a selection, say out loud whether you are using labels or positions, and how many rows you expect. Then check with .shape.