添加与删除列

The simplest way to add a column is direct assignment: df['new_col'] = values. The right-hand side can be a scalar (broadcasts to all rows), a list of the same length, a NumPy array, a Pandas Series (aligned by index), or a computed expression involving existing columns. New columns are added to the right of the DataFrame.

One of the most common DataFrame operations is creating derived features from existing columns. You can compute running totals, ratios, flags, or encode categorical values in a single vectorized expression. These computed columns update automatically when the source columns change if you recompute them, making feature engineering pipelines easy to reproduce.

df.assign(new_col=expression) returns a new DataFrame with the added column without modifying the original. Unlike direct assignment it fits naturally into a method chain. You can add multiple columns in one call, and later column expressions can reference earlier columns defined in the same assign() call using lambda functions.

df.rename(columns={'old': 'new'}) returns a new DataFrame with renamed columns. Pass a dict to rename selected columns without affecting others. You can also rename by function: df.rename(columns=str.upper) uppercases all column names. Always rename before merging DataFrames from different sources to ensure key columns align.

Real-world column names often have spaces, special characters, or inconsistent casing. A common preprocessing step is to normalise them: strip whitespace, convert to lowercase, and replace spaces with underscores. This makes columns accessible via dot notation and avoids quoting issues in SQL-like query strings.

df.insert(loc, column, value) inserts a column at integer position loc, shifting other columns to the right. This is useful when you want the new column to appear next to related columns rather than at the end. It modifies the DataFrame in place, unlike assign().

df.drop(columns=['col1', 'col2']) returns a new DataFrame without those columns. The older syntax df.drop(['col1', 'col2'], axis=1) also works but is less readable. To drop in place, pass inplace=True. Always prefer returning a new DataFrame in pipelines to preserve the original for comparison or rollback.

df.drop(index=['r1', 'r2']) removes rows by label. df.drop([0, 2]) removes rows 0 and 2 from a default integer index. Dropping rows is commonly used to remove known bad records, outliers, or test rows after data cleaning. Prefer boolean indexing for condition-based removal to keep the code readable.

df.pop('col') removes the column from the DataFrame in place and returns it as a Series. This is useful when you need to extract a target column (label) from a feature matrix — you remove it from the DataFrame and hold it separately for use as the y variable in a model.

To reorder columns, select them in the desired order using df[['col3', 'col1', 'col2']]. For large DataFrames where you only want to move a few columns, compute the desired order programmatically: move priority columns to the front and let the rest follow. This is a common reporting step before exporting to Excel.

Updating all values of an existing column uses the same assignment syntax as adding a new one: df['col'] = new_values. Use df.loc[mask, 'col'] = value to update only the rows matching a condition. Never use chained indexing for updates — always use .loc to avoid the SettingWithCopyWarning.

Test your understanding of adding and dropping columns from this lesson.

In this lesson you learned: new columns are added via assignment or the method-chain-friendly assign(), rename() changes column names without altering data, and drop() removes columns or rows and returns a new DataFrame by default. Next up we use head(), tail(), info(), and describe() to quickly profile any DataFrame.