How do I join text from several columns or rows in pandas?

The short version

To combine columns row by row, add them with + or use str.cat. To collapse a whole column into a single string, use str.cat(sep=...). Missing values are the trap: they turn the whole combined value into NaN unless you handle them.

Combining columns row by row

import pandas as pd

df = pd.DataFrame({"first": ["Ana", "Bo"], "last": ["Smith", "Jones"]})

df["full"] = df["first"] + " " + df["last"]
print(df["full"].tolist())   # ['Ana Smith', 'Bo Jones']

Simple addition works because pandas applies it element by element to text columns.

The missing-value trap

df.loc[0, "last"] = None
df["first"] + " " + df["last"]     # ['nan', 'Bo Jones'] -> actually NaN for row 0

Adding anything to a missing value gives a missing value, so one gap poisons the whole combined string. Fill the gaps first.

df["full"] = df["first"].fillna("") + " " + df["last"].fillna("")

Or use str.cat, which takes an explicit instruction about what to do with missing values.

df["full"] = df["first"].str.cat(df["last"], sep=" ", na_rep="")

na_rep substitutes a placeholder for missing entries rather than propagating NaN, which is usually what you want and is clearer than a chain of fillna calls.

Joining a whole column into one string

pd.Series(["a", "b"]).str.cat(sep="-")   # 'a-b'

With no other column passed, str.cat collapses the Series into a single string. This is the pandas equivalent of "-".join(...), and it is useful for building a summary line or an SQL fragment.

Joining within groups

The common real need is collecting each group's values into one string.

df.groupby("city")["product"].agg(lambda s: ", ".join(s.dropna()))

The dropna() matters, because join raises on a missing value. This produces one row per city with its products listed, which is a readable summary format.

Numbers need converting first

You cannot add a number to a string. Convert explicitly.

df["label"] = df["city"] + " (" + df["amount"].astype(str) + ")"

Forgetting the astype(str) gives a TypeError, which is at least a loud failure rather than a silent one.

Prefer building strings at the end

A general point: keep numbers as numbers, dates as dates, and build display strings only at the final presentation step. Storing a combined text label in your working frame is fine, but never replace a numeric column with its formatted text version, because every subsequent calculation on it will then fail or, worse, do something strange.