\For in Algorithm environment causing code to not compile

Question

In the code below, when I use the \For statement, the code does not compile and gives the message:

! Missing \endcsname inserted. <to be read again> \ALG@currentbkock@0

Can you help me get the code to compile? Thanks!

Here is the code:

\documentclass{article}
\usepackage{amsmath, amsfonts, amssymb, amsthm, bm}
\usepackage{commath}
\usepackage{xcolor}
\usepackage{algpseudocode}
\usepackage{pifont}
\usepackage[linesnumbered,ruled,vlined]{algorithm2e}
\newcommand\mycommfont[1]{\footnotesize\ttfamily\textcolor{blue}{#1}}
\SetCommentSty{mycommfont}

\begin{document}

\begin{algorithm}[H]
\DontPrintSemicolon
%  \KwData{Training set $x$}
%  $\Delta_{ji}^l := 0$ \tcp*{will be used to compute $\partial x$}
%  \tcc{iterate over all training examples}
{\bfseries Input:} {a matrix of training samples $A = [A_{1}, A_{2}...,A_{k} ] \in \mathbb{R}^{m\times n}$ for $k$ classes, a test sample $\mathbf{y}\in \mathbb{R}^{m}$, (and an optional error tolerance $\varepsilon > 0$).}\\
Normalize the columns of $A$ to have unit $\ell^{2}$-norm.\\
Solve the $\ell^{1}$-minimization problem:
\begin{equation}
\hat{\bm{x}}_{1} = \arg \min_{x}\norm{\bm{x}}_{1}\quad \text{subject to}\quad A\bm{x} = \bm{y}
\end{equation}
(Or alternatively, solve
\begin{equation}
\hat{\bm{x}}_{1} = \arg \min_{x}\norm{\bm{x}}_{1}\quad \text{subject to}\quad \norm{A\bm{x} = \bm{y}}_{2}\leqslant \varepsilon).
\end{equation}\\
Compute the residuals $r_{i}(\bm{y}) = \norm{\bm{y} - A \delta_{i} (\hat{\bm{x}}_{1})}_{2}$\\
\For{$i = 1,\ldots,k$.}
\EndFor\\
{\bfseries Output:} identity$(\bm{y}) = \arg \min_{i}r_{i}$
\caption{Sparse Representation-based Classification (SRC)}
\end{algorithm}

\end{document} 

Answer 1

The problem is as a result of you mixing two incompatible algorithm packages. algorithm2e provides both the algorithm environment as well as the syntax macros for writing pseudocode. algpseudocode provides only syntax macros. You should use either algorithm2e or algpseudocode (and algorithm), but not both.

Here is a rough algorithm2e implementation of what you might be after:

\documentclass{article}
\usepackage{amsmath,amssymb,commath,bm}
\usepackage[linesnumbered,ruled,vlined]{algorithm2e}

\newcommand\mycommfont[1]{\footnotesize\ttfamily\textcolor{blue}{#1}}
\SetCommentSty{mycommfont}

% http://tex.stackexchange.com/a/153906/5764
\let\oldnl\nl% Store \nl in \oldnl
\newcommand{\nonl}{\renewcommand{\nl}{\let\nl\oldnl}}% Remove line number for one line

\begin{document}

\begin{algorithm}[H]
  \DontPrintSemicolon
  \caption{Sparse Representation-based Classification (SRC)}
  \KwIn{a matrix of training samples $A = [A_{1}, A_{2}, \dots ,A_{k}] \in \mathbb{R}^{m \times n}$ 
    for $k$ classes, a test sample $\mathbf{y} \in \mathbb{R}^{m}$, (and an optional error tolerance $\varepsilon > 0$).}

  Normalize the columns of $A$ to have unit $\ell^{2}$-norm.\;
  Solve the $\ell^{1}$-minimization problem:
  $\hat{\bm{x}}_{1} = \arg \min_{x}\norm{\bm{x}}_{1}\quad \text{subject to}\quad A\bm{x} = \bm{y}$ \;
  \nonl (Or alternatively, solve
  $\hat{\bm{x}}_{1} = \arg \min_{x}\norm{\bm{x}}_{1}\quad \text{subject to}\quad \norm{A\bm{x} = \bm{y}}_{2} \leqslant \varepsilon$).\;
  Compute the residuals $r_{i}(\bm{y}) = \norm{\bm{y} - A \delta_{i}(\hat{\bm{x}}_{1})}_{2}$\;
  \For{$i = 1,\dots,k$}{something}
\end{algorithm}

\end{document}