Level 1 · Rows by position, columns by name

Two ways to point at rows

Every DataFrame row has a position (0 for the first row, 1 for the second, and so on) and an index label (the value shown on the left). Right after loading a CSV they are the same numbers, which hides the difference. After filtering or sorting, they’re not, and picking the wrong one gives you the wrong rows without any error.

pandas gives you one tool for each:

• .iloc selects by integer position.
• .loc selects by label: index labels for rows, column names for columns.

iloc: positions

.iloc works like Python list slicing, end excluded:

products.iloc[0]          # first row, as a Series
products.iloc[10:15]      # positions 10, 11, 12, 13, 14
products.iloc[-3:]        # last three rows
products.iloc[10:15, 1:4] # same rows, columns at positions 1 to 3

Column positions are fragile, because adding a column shifts them. So in practice you’ll usually pick rows with .iloc and columns by name in a second step:

products.iloc[10:15][["name", "category", "unit_price"]]

loc: labels

.loc[rows, columns] takes labels for both:

products.loc[10:14, ["name", "category", "unit_price"]]
products.loc[:, ["name", "unit_price"]]     # all rows, two columns
products.loc[3, "name"]                     # one cell

Important: label slices in .loc include the end. products.loc[10:14] returns five rows, labels 10 through 14. That’s because labels aren’t always numbers (they can be dates or names), and “from Monday to Friday” should include Friday.

When labels and positions drift apart

Sort the products by price and the index labels travel with their rows:

by_price = products.sort_values("unit_price", ascending=False)
by_price.iloc[0]       # the most expensive product (position 0)
by_price.loc[0]        # the row whose label is 0, wherever it ended up

If you meant “the first row”, iloc is right and loc silently gives you a different product. That’s why you shouldn’t rely on index labels after reshaping the data unless you set them on purpose, for example with reset_index(drop=True) or set_index("product_id").

Using a meaningful index

You can make an ID column the index, then look rows up by ID:

by_id = products.set_index("product_id")
by_id.loc[102, "name"]        # 'Alder Oak Coffee Table'
by_id.loc[[101, 102, 103]]    # several products at once

This is the pandas version of a dictionary lookup, and it’s handy when you need the same rows repeatedly.

The one-step rule

When you want to change values, do it in one .loc call: df.loc[row_selector, "column"] = value. Chaining two selections, such as df[df["x"] > 0]["column"] = value, may modify a temporary copy instead of df. Modern pandas warns about this (and pandas 3 never modifies the original that way), so form the habit now.

Common mistakes

• Expecting products.loc[10:15] to give five rows. It gives six, because .loc includes the end.
• Using .loc[0] after a sort to get the top row.
• Mixing positions and names in one .iloc call: products.iloc[10:15, "name"] fails, since .iloc only accepts integers.