Level 2 · Only the marketplace orders

A comparison on a column gives a mask

Compare a column with a value and pandas compares every row at once:

orders["channel"] == "marketplace"

0        False
1        False
2         True
3         True
...
Name: channel, Length: 12000, dtype: bool

The result is a Series of booleans, one per row, called a boolean mask. It has the same index as orders, so each True or False lines up with its row.

All the comparison operators work this way: ==, !=, <, <=, >, >=.

orders["shipping_fee"] > 0
products["is_active"] == False
order_items["quantity"] >= 5

Using the mask

Put a mask inside square brackets and pandas keeps the rows where it’s True:

marketplace = orders[orders["channel"] == "marketplace"]
len(marketplace)      # 1767

You can also name the mask first. For anything beyond one condition, that’s easier to read and debug:

is_marketplace = orders["channel"] == "marketplace"
is_marketplace.sum()          # 1767: True counts as 1
is_marketplace.mean()         # 0.147: the share of orders
orders[is_marketplace]

That .sum() and .mean() trick is worth remembering. Counting how many rows match a condition, or what share do, doesn’t need a filter at all.

The same mask works with .loc, which also lets you pick columns:

orders.loc[is_marketplace, ["order_id", "order_date", "shipping_fee"]]

The index survives

Filtered rows keep their original index labels, so the result’s index has gaps (2, 3, 9, 15…). That’s normal and often useful, since you can trace each row back to the source. When you want a clean 0, 1, 2 index for display or export, add .reset_index(drop=True). Without drop=True, the old index becomes a new column called index, which is rarely what you want.

Exact text matching

Comparisons on text are exact and case-sensitive. orders["channel"] == "Marketplace" matches nothing, and pandas won’t warn you: you just get an empty DataFrame. If you’re filtering on a value you haven’t seen, check the real values first with orders["channel"].unique().

Filtering never changes the original

orders[mask] returns a new DataFrame. orders still has all 12,000 rows. If you plan to modify the filtered result (add a column, fix values), take a copy: marketplace = orders[mask].copy().

Common mistakes

• Writing orders["channel"] = "marketplace" with one equals sign. That doesn’t filter; it overwrites the whole channel column.
• Getting an empty result because of casing or extra spaces in the comparison value.
• Forgetting the column: orders[orders == "marketplace"] compares every cell and returns a table full of NaN.

If you know SQL

orders[orders["channel"] == "marketplace"] is SELECT * FROM orders WHERE channel = 'marketplace'.