melt: de formato ancho a formato largo

Data can be structured in two fundamental shapes. Wide format has one row per subject and spreads measurements across multiple columns — for example, separate columns for January, February, and March sales. Long format has one row per observation, with a column for the variable name and another for the value. Most Pandas analysis functions, machine learning libraries, and visualisation tools prefer long format, making melt() an essential transformation tool.

pd.melt(df) (also available as df.melt()) converts a wide DataFrame to long format. The key parameters are id_vars (columns to keep as identifiers, unchanged), value_vars (columns to unpivot into rows), var_name (name for the new variable column), and value_name (name for the new value column).

Call melt() with id_vars set to the columns you want to keep as identifiers. Every other column is unpivoted: its name becomes a value in the new variable column and its cell values move to the value column. The number of output rows equals the number of input rows times the number of value columns being melted.

By default, every column not listed in id_vars is melted. Use value_vars to melt only a subset of columns. Columns not in id_vars or value_vars are dropped from the result. This is useful when your wide DataFrame contains a mix of time-series columns and other attributes you do not want to unpivot.

melt() preserves the original row indices by repeating them for each variable. The result often has a non-sequential index like [0, 1, 0, 1, 0, 1]. It is good practice to call reset_index(drop=True) immediately after melt() to get a clean sequential index from 0 to n-1 in the output.

melt() is conceptually the reverse of pivot() / pivot_table(). You can go from long to wide with pivot_table() and from wide back to long with melt(). This round-trip is often used in pipelines: receive data in wide format, reshape to long for Seaborn plotting or ML features, then optionally pivot back for a report table.

Seaborn's categorical and relational plots work best with long-format data. A typical workflow: start with a wide DataFrame that has a column per group, melt to long format so that one column identifies the group and another holds the values, then pass both to Seaborn's hue and x/y parameters. This is one of the most common reasons to use melt().

By default, melt() names the new variable column 'variable' and the value column 'value'. Always override these with meaningful names using var_name and value_name. Descriptive column names make downstream code easier to read and prevent confusion when a DataFrame has dozens of columns.

A classic use case for melt() is survey responses where each column is a question and each row is a respondent. Melting converts this wide format into long format with columns for respondent ID, question name, and answer value. You can then easily compute the distribution of answers for each question using groupby().

After melting, the rows are ordered by the original column order (Jan, Feb, Mar for each row). You often want to sort by the identifier column first, then by the variable column. Use sort_values() on the melted DataFrame to achieve the order that makes most sense for your analysis or downstream processing.

In a typical data pipeline, melt() appears right after loading or cleaning wide-format data, before any analysis or modelling step. Place it early in your pipeline to get to long format quickly, then all subsequent operations (groupby, seaborn, sklearn) work on the clean long-format data without special-casing.

Test your understanding of pd.melt wide-to-long transformation from this lesson.

In this lesson you learned: melt() converts wide to long format by unpivoting columns into rows; id_vars specifies which columns stay fixed and value_vars selects which columns to melt; var_name and value_name give descriptive names to the new columns; and long format is preferred for Seaborn, groupby, and ML pipelines. Next up we explore stack and unstack for reshaping MultiIndex DataFrames.