Algorithms and source listings

Papers present code in two distinct forms, and LaTeX has a separate tool for each. Pseudocode is an idealized sketch of an algorithm, typeset with keywords and math, and the algorithm2e package handles it. Source listings are real code reproduced verbatim in a monospaced font, and the listings package handles those. Choosing the wrong tool is the usual source of friction, so it helps to see both.

Pseudocode with algorithm2e

\usepackage[ruled,vlined]{algorithm2e}
\begin{algorithm}
\caption{Train}
\While{not converged}{
 Update $\theta$\;
}
\end{algorithm}

The package options set the visual style: ruled draws horizontal rules above and below the algorithm and places the caption at the top, the convention most venues expect, while vlined draws a small vertical line marking the extent of each block, so nesting stays readable without heavy indentation. The algorithm environment is a float, like a figure, so it carries the \caption that gives it a number for \ref to cite.

Inside, \While{condition}{body} typesets the loop with a bold keyword, the condition, and the body as an indented block; \For, \If, and \Return follow the same pattern. Statements end with \;, which prints the line’s terminator and starts a new line, and forgetting it is the classic beginner error, since the lines then run together. Because this is ordinary LaTeX, math like $\theta$ works inline, which is precisely what pseudocode is for: describing the update rule in notation rather than in any particular language.

Real code with listings

\usepackage{listings}
\begin{lstlisting}[language=Python]
def f(x):
 return x**2
\end{lstlisting}

The lstlisting environment reproduces its contents exactly, preserving whitespace and treating special characters literally, so you can paste code without escaping anything. The language=Python option turns on keyword highlighting for that language, and listings knows dozens of others. By default the output is plain; most people add a one-time \lstset{basicstyle=\ttfamily\small, numbers=left} in the preamble to get a monospaced font at a sensible size with line numbers. To include a file directly instead of pasting, \lstinputlisting[language=Python]{train.py} reads it from disk, which keeps the paper in sync with the actual source. If you want colored, high-fidelity highlighting, the minted package is the stronger alternative, at the cost of requiring an external tool and shell escape.

Choosing between them

Use pseudocode in the paper body to communicate the idea, and reserve verbatim listings for artifacts where exact code matters, such as appendices or tool descriptions. A common mistake is placing a lstlisting inside a Beamer frame without the [fragile] option, which fails with an opaque error; the same caution appears in a first Beamer deck.