Python Pandas Tutorial (Part 6): Add/Remove Rows and Columns From DataFrames

Adding and removing data in pandas DataFrames comes down to a few core operations: assigning new columns from computed Series, using drop to delete columns or rows, and using append (plus ignore_index) to grow a table by rows. The most practical takeaway is that column changes are usually done by assignment (with bracket notation), while row changes often require either append for growth or drop for deletion—then you decide whether to keep changes in place.

To add a column, pandas expects a Series-like object with the same length as the DataFrame. A common example combines two existing string columns—first name and last name—into a single new column. The workflow is to create a combined Series (e.g., first_name + " " + last_name) and then assign it to a new column name using bracket notation: df["full name"] = combined_series. That bracket requirement matters: dot notation (df.full_name = ...) is treated as an attribute assignment on the DataFrame object rather than a column operation, so bracket notation is the reliable approach. For more complex transformations, the same pattern can be paired with apply, letting one column’s values drive new computed columns.

Removing columns uses df.drop with a list of column labels. By default, drop returns a new DataFrame view of what the result would look like; to make the change permanent, set in_place=True. The tutorial also demonstrates reversing a transformation: splitting a combined “full name” column back into “first” and “last.” Using the string split method on the Series produces lists per row; setting expand=True turns those lists into separate columns. Those split outputs then get assigned back into two new DataFrame columns via multi-column indexing (df[["first", "last"]] = split_result).

Row operations follow a different pattern. Adding a single row uses append with a dictionary or Series of values, but pandas requires index handling: if the appended object lacks a name/index, ignore_index=True prevents errors and forces pandas to assign the next index automatically. Appending an entire DataFrame works similarly: df.append(df2, ignore_index=True) stacks rows from df2 onto df, even when indexes conflict. If pandas warns about column ordering, passing sort=False avoids sorting columns and suppresses the warning.

Removing rows mirrors column deletion but targets indexes instead of labels. Dropping a specific row uses df.drop(index=...) with the row index value(s); again, in_place=True makes the removal stick. For conditional row removal, drop can take an index derived from a boolean condition—such as dropping rows where last name equals a target value—often by computing a filtered index first (e.g., filt = df[df["last name"] == "DOE"].index) and then dropping df.drop(index=filt). The result is cleaner, more readable code than embedding the condition directly inside the drop call.