query, isin and between

Boolean masks are correct and they don't scale to reading well. Three conditions and the frame's name appears six times:

orders[(orders["cups"] > 100) & (orders["city"] == "Delhi") & (orders["rating"] > 4)]

query

.query() takes the condition as a string, with the column names bare:

orders.query("cups > 100")
orders.query("cups > 100 and city == 'Delhi' and rating > 4")

Note you can use and, or and not here. Inside the string they're safe — pandas parses the expression itself rather than letting Python evaluate it, so the ambiguity that forces & in a mask doesn't arise.

Reference a Python variable with @:

threshold = 100
orders.query("cups > @threshold")
orders.query("city in @wanted")

A column name with a space in it needs backticks — orders.query("`unit price` > 20").

isin and between

Two helpers that collapse long chains:

orders[orders["city"].isin(["Delhi", "Pune"])]      # instead of a stack of |
orders[orders["cups"].between(90, 130)]            # inclusive at BOTH ends
orders[orders["cups"].between(90, 130, inclusive="neither")]
orders[~orders["city"].isin(["Delhi"])]            # ~ negates any mask

between including both ends is the opposite of a Python slice, so it's worth checking when a boundary value matters.

Text conditions

The .str accessor gives you vectorised versions of the usual string methods, each returning a mask:

orders[orders["item"].str.contains("ginger")]
orders[orders["item"].str.contains("ginger", case=False)]   # catches "Ginger"
orders[orders["city"].str.startswith("M")]
orders[orders["item"].str.match(r"^gin")]                   # regex

Watch out for missing values: .str.contains returns NaN for them, which then can't be used as a mask. Pass na=False to treat gaps as "no match", which is nearly always what you want.

Which to use

Masks for one or two conditions, and any time you want the mask itself — to count with .sum(), or reuse it. query when there are several and readability wins.

Two honest downsides to query: it's a little slower, since it parses a string on every call; and a typo in a column name isn't caught until the line runs, whereas a mask fails where you wrote it.