This R package translates R objects to suitable elements in MathML or LaTeX, thereby allowing for a pretty mathematical representation of R objects and functions in data analyses, scientific reports and interactive web content. In the R Markdown document rendering language, R code and mathematical content already exist side-by-side. The present package enables use of the same R objects for both data analysis and typesetting in documents or web content. This tightens the link between the statistical analysis and its verbal description or symbolic representation, which is another step towards reproducible science. User-defined hooks enable extension of the package by mapping specific variables or functions to new MathML and LaTeX entities. Throughout the paper, examples are given for the functions of the package, and a case study illustrates its use in a scientific report.
\(\require{cancel}\)
The R extension of the markdown language (Xie et al. 2018; Allaire et al. 2023)
enables reproducible statistical reports with
nice typesetting in HTML, Microsoft Word, and LaTeX. Moreover, since
recently (R Core Team 2022, version 4.2), Rs manual pages include
support for mathematical expressions (Sarkar and Hornik 2022; Viechtbauer 2022),
which is already a big improvement. However, except
for special cases such as regression models (Anderson et al. 2023)
and R’s own plotmath annotation, rules for the mapping of
built-in language elements to their mathematical representation are
still lacking. So far, R expressions such as pbinom(k, N, p) are
printed as they are and pretty mathematical formulae such as
\(P_{\mathrm{Bi}}(X \le k; N, p)\) require explicit LaTeX commands like
P_{\mathrm{Bi}}\left(X \le k; N, p\right). Except for very basic use
cases, these commands are tedious to type and their source code is hard
to read.
The present R package defines a set of rules for the automatic translation of R expressions to mathematical output in R Markdown documents (Xie et al. 2020) and Shiny Apps (Chang et al. 2022). The translation is done by an embedded Prolog interpreter that maps nested expressions recursively to MathML and LaTeX/MathJax, respectively. User-defined hooks enable extension of the set of rules, for example, to represent specific R elements by custom mathematical signs.
The main feature of the package is that the same R expressions and equations can be used for both mathematical typesetting and calculations. This saves time and potentially reduces mistakes, as will be illustrated below. Readers should have basic knowledge of knitr and R Markdown to be able to follow this article (Allaire et al. 2023; Xie 2023), while to extend and customize the package, some basic knowledge of Prolog is needed.
The paper is organized as follows. We start with a description of the technical background of the package, including the two main classes of rules for translating R objects to mathematical expressions. The next section illustrates the main features of the mathml package, potential issues and their workarounds using examples from the day-to-day perspective of a user. A case study follows with a scientific report written with the help of the package. The last section concludes with a discussion and ideas for further development.
Similar to other high-level programming languages, R is homoiconic, that
is, R commands (i.e., R “calls”) are, themselves, symbolic data
structures that can be created, parsed and modified. Because the default
response of the R interpreter is to evaluate a call and return its
result, this property is not transparent to the general user. There
exists, however, a number of built-in R functions (e.g., quote(),
call() etc.) that allow the user to create R calls which can be stored
in regular variables and then, for example, evaluated at a later stage
or in a specific environment (Wickham 2019). The present package
includes a set of rules that translate such calls to a mathematical
representation in MathML and LaTeX. For a first illustration of the
mathml package, we consider the binomial probability.
term <- quote(pbinom(k, N, p))The term is quoted to avoid its immediate evaluation (which would raise
an error anyway since the variables k, N, p have not yet been
defined). Experienced R users will recognize that the expression is a
short form for
term <- call("pbinom", as.name("k"), as.name("N"), as.name("p"))
term
#> pbinom(k, N, p)As can be seen from the output, to the variable term is not assigned
the result of the calculation, but instead an R call (see, e.g., Wickham 2019,
for details on “non-standard evaluation”), which can eventually be
evaluated with eval(),
k <- 10
N <- 22
p <- 0.4
eval(term)
#> [1] 0.77195The R package mathml can now be used to render the call in MathML or in MathJax/LaTeX. MathML is the dialect for mathematical elements on HTML webpages, whereas LaTeX is typically used for typesetting printed documents, as shown below.
library(mathml)
substr(mathml(term), 1, 70)
#> [1] "<math><mrow><msub><mi>P</mi><mtext>Bi</mtext></msub><mo>⁡</mo><mrow"
mathjax(term)
#> [1] "${P}_{\\mathrm{Bi}}{\\left({{X}{\\le}{k}}{{;}{{N}{{,}{p}}}}\\right)}$"Some of the curly braces are not really needed in the LaTeX output, but
are necessary in edge cases. The package also includes a function
mathout() that wraps a call to mathml() for HTML output and
mathjax() for LaTeX output. Moreover, the function math(x) adds the
class "math" to its argument, such that a special knitr printing
function is invoked (see the vignette on custom print methods in Xie 2023).
An R Markdown code chunk with mathout(term) thus produces:
\({P}_{\mathrm{Bi}}{\left({{X}{\le}{k}}{{;}{{N}{{,}{p}}}}\right)}\)
Similarly, inline() produces inline output, r inline(term) yields
\({P}_{\mathrm{Bi}}{\left({{X}{\le}{k}}{{;}{{N}{{,}{p}}}}\right)}\).
The currently supported R objects are listed below, roughly following the order proposed by Murrell and Ihaka (2000).
mathml handles the basic elements of everyday mathematical expressions, such as integers, floating-point numbers, Latin and Greek letters, multi-letter identifiers, accents, subscripts, and superscripts.
term <- quote(1 + -2L + a + abc + "a" + phi + Phi + varphi + roof(b)[i, j]^2L)
math(term)\({{{{{{{{1.00}{+}{{-}{2}}}{+}{a}}{+}{abc}}{+}{\mathrm{a}}}{+}{\phi}}{+}{\Phi}}{+}{\varphi}}{+}{{\hat{b}}_{{i}{{\mathrm{}}{j}}}^{2}}\)
term <- quote(round(3.1415, 3L) + NaN + NA + TRUE + FALSE + Inf + (-Inf))
math(term)\({{{{{{3.142}{+}{nan}}{+}{na}}{+}{T}}{+}{F}}{+}{\infty}}{+}{\left({-}{\infty}\right)}\)
An expression such as 1 + -2 may be considered aesthetically
unsatisfactory. It is correct R syntax, though, and is reproduced
accordingly, without the parentheses. Parentheses around negative
numbers or symbols can be added as shown above for + (-Inf).
To avoid name clashes with package stats, roof() is used to put a
hat on a symbol (see next section for further decorations). Note that an
R function roof() does not exist in base R, it is provided by the
package for convenience and points to the identity function.
The package offers some support for different fonts as well as accents and boxes etc. Internally, these decorations are implemented as identity functions, so that they can be introduced into R expressions without side-effects.
term <- quote(bold(b[x, 5L]) + bold(b[italic(x)]) + italic(ab) + italic(42L))
math(term)\({{{{\mathbf{b}}_{{\mathbf{x}}{{\mathrm{}}{5}}}}{+}{{\mathbf{b}}_{\mathit{x}}}}{+}{\mathit{ab}}}{+}{42}\)
term <- quote(tilde(a) + mean(X) + box(c) + cancel(d) + phantom(e) + prime(f))
math(term)\({{{{{\tilde{a}}{+}{\overline{X}}}{+}{\boxed{c}}}{+}{\cancel{d}}}{+}{\phantom{e}}}{+}{{f^\prime}}\)
Note that the font styles only affect the display of identifiers, whereas numbers, character strings etc. are left untouched.
Arithmetic operators and parentheses are translated as they are, as illustrated below.
term <- quote(a - ((b + c)) - d*e + f*(g + h) + i/j + k^(l + m) + (n*o)^{p + q})
math(term)\({{{{{{a}{-}{\left[\left({b}{+}{c}\right)\right]}}{-}{{d}{{}}{e}}}{+}{{f}{\cdot}{\left({g}{+}{h}\right)}}}{+}{{i}{/}{j}}}{+}{{k}^{\left({l}{+}{m}\right)}}}{+}{{\left({n}{{}}{o}\right)}^{{p}{+}{q}}}\)
term <- quote(dot(a, b) + frac(1L, nodot(c, d + e)) + dfrac(1L, times(g, h)))
math(term)\({{{a}{\cdot}{b}}{+}{\frac{1}{{c}{{}}{\left({d}{+}{e}\right)}}}}{+}{\displaystyle{\frac{1}{{g}{\times}{h}}}}\)
For multiplications involving only numbers and symbols, the
multiplication sign is omitted. This heuristic does not always produce
the desired result; therefore, mathml defines alternative R functions
dot(), nodot(), and times(). These functions calculate a product
and produce the respective multiplication signs. Similarly, frac() and
dfrac() can be used for small and large fractions.
For standard operators with known precedence, mathml is generally able
to detect if parentheses are needed; for example, parentheses are
automatically placed around d + e in the nodot-example. However, we
note unnecessary parentheses around l + m above. These parentheses are
a consequence of quote(a^(b + c)) actually producing a nested R call
of the form ^(a, (b + c)) instead of ^(a, b + c):
term <- quote(a^(b + c))
paste(term)
#> [1] "^" "a" "(b + c)"For the present purpose, this R feature is unfortunate because the extra
parentheses around b + c are not needed. The preferred result is
obtained by the functional form quote(^(k, l + m)) of the power, or
curly braces as a workaround (see p + q above).
Whereas in standard infix operators, the parentheses typically follow the rules for precedence, undesirable results may be obtained in custom operators.
term <- quote(mean(X) %+-% 1.96 * s / sqrt(N))
math(term)\({{\left({\overline{X}}{\pm}{1.96}\right)}{\cdot}{s}}{/}{\sqrt{N}}\)
term <- quote('%+-%'(mean(X), 1.96 * s / sqrt(N))) # functional form of '%+-%'
term <- quote(mean(X) %+-% {1.96 * s / sqrt(N)}) # the same
math(term)\({\overline{X}}{\pm}{{{1.96}{{}}{s}}{/}{\sqrt{N}}}\)
The example is a reminder that it is not possible to define the precedence of custom operators in R, and that expressions with such operators are evaluated strictly from left to right. Again, the problem can be worked around by the functional form of the operator or a curly brace to hide the parenthesis, and, at the same time, enforce the correct operator precedence.
More operators are shown in Table 1, including the suggestions by Murrell and Ihaka (2000) for graphical annotations and arrows in R figures.
| Operator | Output | Operator | Output | Operator | Arrow |
|---|---|---|---|---|---|
| A %*% B | \({A}{\times}{B}\) | A != B | \({A}{\ne}{B}\) | A %% B | \({A}{\leftrightarrow}{B}\) |
| A %.% B | \({A}{\cdot}{B}\) | A ~ B | \({A}{\sim}{B}\) | A %->% B | \({A}{\rightarrow}{B}\) |
| A %x% B | \({A}{\otimes}{B}\) | A %~~% B | \({A}{\approx}{B}\) | A %<-% B | \({A}{\leftarrow}{B}\) |
| A %/% B | \(\lfloor{{A}{/}{B}}\rfloor\) | A %==% B | \({A}{\equiv}{B}\) | A %up% B | \({A}{\uparrow}{B}\) |
| A %% B | \(mod{\left({A}{{,}{B}}\right)}\) | A %=~% B | \({A}{\cong}{B}\) | A %down% B | \({A}{\downarrow}{B}\) |
| A & B | \({A}{\land}{B}\) | A %prop% B | \({A}{\propto}{B}\) | A %<=>% B | \({A}{\iff}{B}\) |
| A | B | \({A}{\lor}{B}\) | A %in% B | \({A}{\in}{B}\) | A %=>% B | \({A}{\Rightarrow}{B}\) |
| xor(A, B) | \({A}{\veebar}{B}\) | intersect(A, B) | \({A}{\cap}{B}\) | A %<=% B | \({A}{\Leftarrow}{B}\) |
| !A | \({\lnot}{A}\) | union(A, B) | \({A}{\cup}{B}\) | A %dblup% B | \({A}{\Uparrow}{B}\) |
| A == B | \({A}{=}{B}\) | crossprod(A, B) | \({{A}^{\mathrm{T}}}{\times}{B}\) | A %dbldown% B | \({A}{\Downarrow}{B}\) |
| A <- B | \({A}{=}{B}\) | is.null(A) | \({A}{=}{\emptyset}\) | \(\mathrm{}\) |
| Function | Output | Function | Output |
|---|---|---|---|
| sin(x) | \(\sin{x}\) | dbinom(k, N, pi) | \({P}_{\mathrm{Bi}}{\left({{X}{=}{k}}{{;}{{N}{{,}{\pi}}}}\right)}\) |
| cosh(x) | \(\cosh{x}\) | pbinom(k, N, pi) | \({P}_{\mathrm{Bi}}{\left({{X}{\le}{k}}{{;}{{N}{{,}{\pi}}}}\right)}\) |
| tanpi(alpha) | \(\tan{\left({\alpha}{{}}{\pi}\right)}\) | qbinom(p, N, pi) | \({\arg\min}_{k}{\left[{{P}_{\mathrm{Bi}}{\left({{X}{\le}{k}}{{;}{{N}{{,}{\pi}}}}\right)}}{>}{p}\right]}\) |
| asinh(x) | \({\sinh}^{{-}{1}}{x}\) | dpois(k, lambda) | \({P}_{\mathrm{Po}}{\left({{X}{=}{k}}{{;}{\lambda}}\right)}\) |
| log(p) | \(\log{p}\) | ppois(k, lambda) | \({P}_{\mathrm{Po}}{\left({{X}{\le}{k}}{{;}{\lambda}}\right)}\) |
| log1p(x) | \(\log{\left({1}{+}{x}\right)}\) | qpois(p, lambda) | \({{\arg\max}}_{k}{\left[{{P}_{\mathrm{Po}}{\left({{X}{\le}{k}}{{;}{\lambda}}\right)}}{>}{p}\right]}\) |
| logb(x, e) | \({\log}_{e}{x}\) | dexp(x, lambda) | \({f}_{\mathrm{Exp}}{\left({x}{{;}{\lambda}}\right)}\) |
| exp(x) | \(\exp{x}\) | pexp(x, lambda) | \({F}_{\mathrm{Exp}}{\left({x}{{;}{\lambda}}\right)}\) |
| expm1(x) | \({\exp{x}}{-}{1}\) | qexp(p, lambda) | \({F}_{\mathrm{Exp}}^{{-}{1}}{\left({p}{{;}{\lambda}}\right)}\) |
| choose(n, k) | \(\binom{n}{k}\) | dnorm(x, mu, sigma) | \(\phi{\left({x}{{;}{{\mu}{{,}{\sigma}}}}\right)}\) |
| lchoose(n, k) | \(\log{\binom{n}{k}}\) | pnorm(x, mu, sigma) | \(\Phi{\left({x}{{;}{{\mu}{{,}{\sigma}}}}\right)}\) |
| factorial(n) | \({n}{!}\) | qnorm(alpha/2L) | \({\Phi}^{{-}{1}}{\left({\alpha}{/}{2}\right)}\) |
| lfactorial(n) | \(\log{{n}{!}}\) | 1L - pchisq(x, 1L) | \({1}{-}{{F}_{{\chi}^{2}{\left({1}{{\,}{\mathrm{df}}}\right)}}{\left(x\right)}}\) |
| sqrt(x) | \(\sqrt{x}\) | qchisq(1L - alpha, 1L) | \({F}_{{\chi}^{2}{\left({1}{{\,}{\mathrm{df}}}\right)}}^{{-}{1}}{\left({1}{-}{\alpha}\right)}\) |
| mean(X) | \(\overline{X}\) | pt(t, N - 1L) | \(P{\left({{T}{\le}{t}}{{;}{{{N}{-}{1}}{{\,}{\mathrm{df}}}}}\right)}\) |
| abs(x) | \({\left\vert{x}\right\vert}\) | qt(alpha/2L, N - 1L) | \({T}_{{\alpha}{/}{2}}{\left({{N}{-}{1}}{{\,}{\mathrm{df}}}\right)}\) |
There is support for most functions from package base, with adequate
use and omission of parentheses.
term <- quote(sin(x) + sin(x)^2L + cos(pi/2L) + tan(2L*pi) * expm1(x))
math(term)\({{{\sin{x}}{+}{{\left(\sin{x}\right)}^{2}}}{+}{\cos{\left({\pi}{/}{2}\right)}}}{+}{{\tan{\left({2}{{}}{\pi}\right)}}{\cdot}{\left({\exp{x}}{-}{1}\right)}}\)
term <- quote(choose(N, k) + abs(x) + sqrt(x) + floor(x) + exp(frac(x, y)))
math(term)\({{{{\binom{N}{k}}{+}{{\left\vert{x}\right\vert}}}{+}{\sqrt{x}}}{+}{\lfloor{x}\rfloor}}{+}{\exp{\left(\frac{x}{y}\right)}}\)
A few more examples are shown in Table
2,
including functions from stats.
For self-written functions, the matter is somewhat more complicated. For
instance, if we consider a function such as g <- function(...) ...,
the name g is not transparent to R, because only the function body
is represented. We can still display functions in the form head(x) = body
if we embed the object to be shown into a call "<-"(head, body).
sgn <- function(x)
{
if(x == 0L) return(0L)
if(x < 0L) return(-1L)
if(x > 0L) return(1L)
}
math(sgn)\(\left\{\begin{array}{l}{{0},\ \mathrm{if}\ {{x}{=}{0}}}\\ {{{-}{1}},\ \mathrm{if}\ {{x}{<}{0}}}\\ {{1},\ \mathrm{if}\ {{x}{>}{0}}}\end{array}\right.\)
math(call("<-", quote(sgn(x)), sgn))\({\mathrm{sgn}\,{x}}{=}{\left\{\begin{array}{l}{{0},\ \mathrm{if}\ {{x}{=}{0}}}\\ {{{-}{1}},\ \mathrm{if}\ {{x}{<}{0}}}\\ {{1},\ \mathrm{if}\ {{x}{>}{0}}}\end{array}\right.}\)
The function body is generally a nested R call of the form {(L), with
L being a list of commands (the semicolon, not necessary in R, is
translated to a newline). The example also illustrates that mathml
provides limited support for control structures such as if that is
internally represented as if(condition, action).
Indices in square brackets are rendered as subscripts, powers are
rendered as superscript. Moreover, mathml defines the functions
sum_over(x, from, to), and prod_over(x, from, to) that simply return
their first argument. The other two arguments serve as decorations (to
is optional), for example, for summation and product signs.
term <- quote(S[Y]^2L <- frac(1L, N) * sum(Y[i] - mean(Y))^2L)
math(term)\({{S}_{Y}^{2}}{=}{{\frac{1}{N}}{\cdot}{{\sum{\left({{Y}_{i}}{-}{\overline{Y}}\right)}}^{2}}}\)
term <- quote(log(prod_over(L[i], i==1L, N)) <- sum_over(log(L[i]), i==1L, N))
math(term)\({\log{{\prod}_{{i}{=}{1}}^{N}{{L}_{i}}}}{=}{{\sum}_{{i}{=}{1}}^{N}{\log{{L}_{i}}}}\)
Rs integrate function takes a number of arguments, the most important
ones being the function to integrate, and the lower and the upper bound
of the integration.
term <- quote(integrate(sin, 0L, 2L*pi))
math(term)\(\int_{0}^{{2}{{}}{\pi}}{\sin{x}}\,{d{x}}\)
eval(term)
#> 2.221482e-16 with absolute error < 4.4e-14For mathematical typesetting in the form of \(\int f(x)\, dx\), mathml
needs to find out the name of the integration variable. For that
purpose, the underlying Prolog bridge provides a predicate r_eval/2
that calls R from Prolog. This predicate is used to evaluate
formalArgs(args(sin)) and returns the names of the arguments of
sin(), namely, x.
Above, the quoted term is an abbreviation for call("integrate", quote(sin), ...),
with sin being an R symbol, not a function. While the R function
integrate() can handle both symbols and functions, mathml needs the
symbol because it is unable to determine the function name of custom
functions.
One of R’s great features is the possibility to refer to function
arguments by their names, not only by their position in the list of
arguments. At the other end, the Prolog handlers for R calls are rather
rigid, for example, integrate/3 accepts exactly three arguments in a
particular order and without names, that is,
integrate(lower=0L, upper=2L*pi, sin), would not print the desired
result.
To “canonicalize” function calls with named arguments and arguments in
unusual order, mathml provides an auxiliary R function
canonical(f, drop) that reorders the argument list of calls to known
R functions and, if drop=TRUE (which is the default), also removes the
names of the arguments.
term <- quote(integrate(lower=0L, upper=2L*pi, sin))
canonical(term)
#> integrate(sin, 0L, 2L * pi)
math(canonical(term))\(\int_{0}^{{2}{{}}{\pi}}{\sin{x}}\,{d{x}}\)
This function can be used to feed mixtures of partially named and
positional arguments into the renderer. For details, see the R function
match.call().
Of course, mathml also supports matrices and vectors.
v <- 1:3
math(call("t", v))\({\left({1}{{\,}{2}{{\,}{3}}}\right)}^{\mathrm{T}}\)
A <- matrix(data=11:16, nrow=2, ncol=3)
B <- matrix(data=21:26, nrow=2, ncol=3)
term <- call("+", A, B)
math(term)\({\left(\begin{array}{ccc}11 & 13 & 15\\ 12 & 14 & 16\\ \end{array}\right)}{+}{\left(\begin{array}{ccc}21 & 23 & 25\\ 22 & 24 & 26\\ \end{array}\right)}\)
Note that the seemingly more convenient term <- quote(A + B) yields
\(A + B\) in the output—instead of the desired matrix representation.
This behavior is expected because quotation of R calls also quote the
components of the call (here, A and B).
In typical R functions, variable names are typically longer than just single letters, which may yield unsatisfactory results in the mathematical output.
term <- quote(pbinom(successes, Ntotal, prob))
math(term)\({P}_{\mathrm{Bi}}{\left({{X}{\le}{successes}}{{;}{{Ntotal}{{,}{prob}}}}\right)}\)
hook(successes, k)
hook(quote(Ntotal), quote(N), quote=FALSE)
hook(prob, pi)
math(term)\({P}_{\mathrm{Bi}}{\left({{X}{\le}{k}}{{;}{{N}{{,}{\pi}}}}\right)}\)
To improve the situation, mathml provides a simple hook that can be used
to replace elements (e.g., verbose variable names) of the code by
concise mathematical symbols, as illustrated in the example. To simplify
notation, hook() uses non-standard evaluation of its arguments. If the
quote flag of hook() is set to FALSE, the user has to provide the
quoted expressions. Care should be taken to avoid recursive hooks such
as hook(s, s["A"]) that endlessly replace the \(s\) from
\(s_{\mathrm{A}}\) as in \(s_{\mathrm{A}_{\mathrm{A}_{\mathrm{A}\cdots}}}\).
The hooks can also be used for more complex elements such as R calls, with dotted symbols representing Prolog variables.
hook(pbinom(.K, .N, .P), sum_over(dbinom(i, .N, .P), i=0L, .K))
math(term)\({\sum}_{{i}{=}{0}}^{k}{{P}_{\mathrm{Bi}}{\left({{X}{=}{i}}{{;}{{N}{{,}{\pi}}}}\right)}}\)
Further customization requires the assertion of new Prolog rules
math/2, ml/3, jax/3, as shown in the Appendix.
We consider the \(t\)-statistic for independent samples with equal
variance. To avoid clutter in the equation, the pooled variance
\(s^2_{\mathrm{pool}}\) is abbreviated, and a comment is given with the
expression for \(s^2_{\mathrm{pool}}\). For this purpose, mathml provides
a function denote(abbr, expr, info), with expr actually being
evaluated, abbr being rendered, plus a comment of the form “with
expr denoting info”.
hook(m_A, mean(X)["A"]) ; hook(s2_A, s["A"]^2L) ;
hook(n_A, n["A"])
hook(m_B, mean(X)["B"]) ; hook(s2_B, s["B"]^2L)
hook(n_B, n["B"]) ; hook(s2_p, s["pool"]^2L)
term <- quote(t <- dfrac(m_A - m_B,
sqrt(denote(s2_p, frac((n_A - 1L)*s2_A + (n_B - 1L)*s2_B, n_A + n_B - 2L),
"the pooled variance.") * (frac(1L, n_A) + frac(1L, n_B)))))
math(term)\({t}{=}{\displaystyle{\frac{{{\overline{X}}_{\mathrm{A}}}{-}{{\overline{X}}_{\mathrm{B}}}}{\sqrt{{{s}_{\mathrm{pool}}^{2}}{\cdot}{\left({\frac{1}{{n}_{\mathrm{A}}}}{+}{\frac{1}{{n}_{\mathrm{B}}}}\right)}}}}}\), with \({{s}_{\mathrm{pool}}^{2}}{=}{\frac{{{\left({{n}_{\mathrm{A}}}{-}{1}\right)}{\cdot}{{s}_{\mathrm{A}}^{2}}}{+}{{\left({{n}_{\mathrm{B}}}{-}{1}\right)}{\cdot}{{s}_{\mathrm{B}}^{2}}}}{{{{n}_{\mathrm{A}}}{+}{{n}_{\mathrm{B}}}}{-}{2}}}\) denoting the pooled variance.
The term is evaluated below. print() is needed because the return
value of an assignment of the form t <- dfrac(...) is not visible in
R.
m_A <- 1.5; s2_A <- 2.4^2; n_A <- 27; m_B <- 3.9; s2_B <- 2.8^2; n_B <- 20
print(eval(term))
#> [1] -3.157427Consider an educational scenario in which we want to highlight a certain element of a term, for example, that a student has forgotten to subtract the null hypothesis in a \(t\)-ratio:
t <- quote(dfrac(omit_right(mean(D) - mu[0L]), s / sqrt(N)))
math(t, flags=list(error="highlight"))\(\displaystyle{\frac{{\overline{D}}{{\,}{\cancel{{-}{{\,}{{\mu}_{0}}}}}}}{{s}{/}{\sqrt{N}}}}\)
math(t, flags=list(error="fix"))\(\displaystyle{\frac{{\overline{D}}{{\,}{\boxed{{-}{{\,}{{\mu}_{0}}}}}}}{{s}{/}{\sqrt{N}}}}\)
The R function omit_right(a + b) uses non-standard evaluation
techniques (e.g., Wickham 2019) to return only the left part an
operation, and cancels the right part. This may not always be desired,
for example, when illustrating how to fix the mistake.
For this purpose, the functions mathml(), mathjax(), mathout() and
math() have an optional argument flags which is a list with named
elements. In this example, we use this argument to tell mathml how to
render such erroneous expressions using the flag error which can be
“asis”, “highlight”, “fix”, or “ignore”. For more examples, see Table
3.
| Operation | error = asis | highlight | fix | ignore |
|---|---|---|---|---|
| omit_left(a + b) | \(b\) | \({\cancel{{a}{{\,}{+}}}}{{\,}{b}}\) | \({\boxed{{a}{{\,}{+}}}}{{\,}{b}}\) | \({a}{+}{b}\) |
| omit_right(a + b) | \(a\) | \({a}{{\,}{\cancel{{+}{{\,}{b}}}}}\) | \({a}{{\,}{\boxed{{+}{{\,}{b}}}}}\) | \({a}{+}{b}\) |
| list(quote(a), quote(omit(b))) | \({a}{{\,}{\mathrm{}}}\) | \({a}{{\,}{\cancel{b}}}\) | \({a}{{\,}{\boxed{b}}}\) | \({a}{{\,}{b}}\) |
| add_left(a + b) | \({a}{+}{b}\) | \({\boxed{{a}{{\,}{+}}}}{{\,}{b}}\) | \({\cancel{{a}{{\,}{+}}}}{{\,}{b}}\) | \(b\) |
| add_right(a + b) | \({a}{+}{b}\) | \({a}{{\,}{\boxed{{+}{{\,}{b}}}}}\) | \({a}{{\,}{\cancel{{+}{{\,}{b}}}}}\) | \(a\) |
| list(quote(a), quote(add(b))) | \({a}{{\,}{b}}\) | \({a}{{\,}{\boxed{b}}}\) | \({a}{{\,}{\cancel{b}}}\) | \({a}{{\,}{\mathrm{}}}\) |
| instead(a, b) + c | \({a}{+}{c}\) | \({\underbrace{a}_{{\mathrm{instead}}{{\,}{\mathrm{of}}{{\,}{b}}}}}{+}{c}\) | \({\boxed{b}}{+}{c}\) | \({b}{+}{c}\) |
This case study describes a model by Schwarz (1994) from mathematical psychology using the features of package mathml. Schwarz (1994) presents a new explanation of redundancy gains that occur when observers respond to stimuli of different sources, and the same information is presented on two or more channels. In Schwarz’s (1994) model, decision-making builds on a process of noisy accumulation of information over time (e.g., Ratcliff et al. 2016). In redundant stimuli, the model assumes a superposition of channel-specific diffusion processes that eventually reach an absorbing barrier to elicit the response. For a detailed description the reader may refer to the original article.
Schwarz’s (1994) model refers to two stimuli A and B, presented either alone or in combination (AB, redundant stimuli), with the redundant stimuli being presented either simultaneously or with onset asynchrony \(\tau\). The channel activation is described as a two-dimensional Wiener process with drifts \(\mu_i\), variances \(\sigma^2_i\), and initial conditions \(X_i(t = 0) = 0, i = \mathrm{A, B}\). The buildup of channel-specific activation may be correlated with \(\rho_{\mathrm{AB}}\), but we assume \(\rho_{\mathrm{AB}} = 0\) for simplicity.
A response is elicited when the process reaches an absorbing barrier \(c > 0\) for the first time. In single-target trials, the first passages of \(c\) are expected at
ED_single <- function(c, mu)
dfrac(c, mu)
# display as E(D; mu), c is a scaling parameter
hook(ED_single(.C, .Mu), E(`;`(D, .Mu)))
math(call("=", quote(ED_single(c, mu)), ED_single))\({E{\left({D}{;}{\mu}\right)}}{=}{\displaystyle{\frac{c}{\mu}}}\)
One would typically use chunk option echo=FALSE to suppress the R
code.
In redundant stimuli, the activation from the channel-specific diffusion processes adds up, \(X_{\mathrm{AB}}(t) = X_{\mathrm A}(t) + X_{\mathrm B}(t)\), hence the name, superposition. \(X_{\mathrm{AB}}(t)\) is again a Wiener process with drift \(\mu_{\mathrm A} + \mu_{\mathrm B}\) and variance \(\sigma^2_{\mathrm A} + \sigma^2_{\mathrm B}\). For the expected first-passage time, we have
hook(mu_A, mu["A"])
hook(mu_B, mu["B"])
hook(sigma_A, sigma["A"])
hook(sigma_B, sigma["B"])
hook(mu_M, mu["M"])
hook(M, overline(X))
math(call("=", quote(E(D["AB"])), quote(ED_single(c, mu_A + mu_B))))\({E{\left({D}_{\mathrm{AB}}\right)}}{=}{E{\left({D}{;}{{{\mu}_{\mathrm{A}}}{+}{{\mu}_{\mathrm{B}}}}\right)}}\)
For asynchronous stimuli, Schwarz (1994) derived the expected first-passage time as a function of the stimulus onset asynchrony \(\tau\),
ED_async <- function(tau, c, mu_A, sigma_A, mu_B)
{ dfrac(c, mu_A) + (dfrac(1L, mu_A) - dfrac(1L, mu_A + mu_B)) *
((mu_A*tau - c) * pnorm(dfrac(c - mu_A*tau, sqrt(sigma_A^2L*tau)))
- (mu_A*tau + c) * exp(dfrac(2L*c*mu_A, sigma_A^2L))
* pnorm(dfrac(-c - mu_A*tau, sqrt(sigma_A^2L*tau))))
}
hook(ED_async(.Tau, .C, .MA, .SA, .MB), E(`;`(D[.Tau], `,`(.MA, .SA, .MB))))
math(call("=", quote(E(D[tau])), ED_async))\({E{\left({D}_{\tau}\right)}}{=}{{\displaystyle{\frac{c}{{\mu}_{\mathrm{A}}}}}{+}{{\left({\displaystyle{\frac{1}{{\mu}_{\mathrm{A}}}}}{-}{\displaystyle{\frac{1}{{{\mu}_{\mathrm{A}}}{+}{{\mu}_{\mathrm{B}}}}}}\right)}{\cdot}{\left[{{\left({{{\mu}_{\mathrm{A}}}{{}}{\tau}}{-}{c}\right)}{\cdot}{\Phi{\left(\displaystyle{\frac{{c}{-}{{{\mu}_{\mathrm{A}}}{{}}{\tau}}}{\sqrt{{{\sigma}_{\mathrm{A}}^{2}}{{}}{\tau}}}}\right)}}}{-}{{{\left({{{\mu}_{\mathrm{A}}}{{}}{\tau}}{+}{c}\right)}{\cdot}{\exp{\left(\displaystyle{\frac{{{2}{{}}{c}}{{}}{{\mu}_{\mathrm{A}}}}{{\sigma}_{\mathrm{A}}^{2}}}\right)}}}{\cdot}{\Phi{\left(\displaystyle{\frac{{{-}{c}}{-}{{{\mu}_{\mathrm{A}}}{{}}{\tau}}}{\sqrt{{{\sigma}_{\mathrm{A}}^{2}}{{}}{\tau}}}}\right)}}}\right]}}}\)
For negative onset asynchrony (i.e., B before A), the parameters are simply switched.
ED <- function(tau, c, mu_A, sigma_A, mu_B, sigma_B)
{
if(tau == Inf) return(ED_single(c, mu_A))
if(tau == -Inf) return(ED_single(c, mu_B))
if(tau == 0L) return(ED_single(c, mu_A + mu_B))
if(tau > 0L) return(ED_async(tau, c, mu_A, sigma_A, mu_B))
if(tau < 0L) return(ED_async(abs(tau), c, mu_B, sigma_B, mu_A))
}
hook(ED(.Tau, .C, .MA, .SA, .MB, .SB), E(`;`(D[.Tau], `,`(.MA, .SA, .MB, .SB))))
math(call("=", quote(ED(tau, c, mu_A, sigma_A, mu_B, sigma_B)), ED))\({E{\left({{D}_{\tau}}{;}{{{\mu}_{\mathrm{A}}}{{,}{{\sigma}_{\mathrm{A}}}{{,}{{\mu}_{\mathrm{B}}}{{,}{{\sigma}_{\mathrm{B}}}}}}}\right)}}{=}{\left\{\begin{array}{l}{{E{\left({D}{;}{{\mu}_{\mathrm{A}}}\right)}},\ \mathrm{if}\ {{\tau}{=}{\infty}}}\\ {{E{\left({D}{;}{{\mu}_{\mathrm{B}}}\right)}},\ \mathrm{if}\ {{\tau}{=}{{-}{\infty}}}}\\ {{E{\left({D}{;}{{{\mu}_{\mathrm{A}}}{+}{{\mu}_{\mathrm{B}}}}\right)}},\ \mathrm{if}\ {{\tau}{=}{0}}}\\ {{E{\left({{D}_{\tau}}{;}{{{\mu}_{\mathrm{A}}}{{,}{{\sigma}_{\mathrm{A}}}{{,}{{\mu}_{\mathrm{B}}}}}}\right)}},\ \mathrm{if}\ {{\tau}{>}{0}}}\\ {{E{\left({{D}_{{\left\vert{\tau}\right\vert}}}{;}{{{\mu}_{\mathrm{B}}}{{,}{{\sigma}_{\mathrm{B}}}{{,}{{\mu}_{\mathrm{A}}}}}}\right)}},\ \mathrm{if}\ {{\tau}{<}{0}}}\end{array}\right.}\)
The observable response time is assumed to be the sum of \(D\), the time employed to reach the threshold for the decision, and a residual \(M\) denoting other processes such as motor preparation and execution. Correspondingly, the expected response time amounts to
ET <- function(tau, c, mu_A, sigma_A, mu_B, sigma_B, mu_M)
ED(tau, c, mu_A, sigma_A, mu_B, sigma_B) + mu_M
hook(ET(.Tau, .C, .MA, .SA, .MB, .SB, .MM),
E(`;`(T[.Tau], `,`(.MA, .SA, .MB, .SB, .MM))))
math(call("=", quote(E(T[tau])), ET))\({E{\left({T}_{\tau}\right)}}{=}{{E{\left({{D}_{\tau}}{;}{{{\mu}_{\mathrm{A}}}{{,}{{\sigma}_{\mathrm{A}}}{{,}{{\mu}_{\mathrm{B}}}{{,}{{\sigma}_{\mathrm{B}}}}}}}\right)}}{+}{{\mu}_{\mathrm{M}}}}\)
Schwarz (1994) applied the model to data from a redundant signals task (Miller 1986) with 13 onset asynchronies \(0, \pm33, \pm67, \pm100, \pm133, \pm167, \pm\infty\) ms, where \(\tau = 0\) refers to the synchronous condition, and \(\pm\infty\) to the single-target presentations. Each condition was replicated 400 times. The observed mean response times and their standard deviations are given in Table 4.
| \(\tau\) | \(m\) | \(s\) | \(n\) |
|---|---|---|---|
| \({-}{\infty}\) | \(231\) | \(56\) | \(400\) |
| \({-}{167}\) | \(234\) | \(58\) | \(400\) |
| \({-}{133}\) | \(230\) | \(40\) | \(400\) |
| \({-}{100}\) | \(227\) | \(40\) | \(400\) |
| \({-}{67}\) | \(228\) | \(32\) | \(400\) |
| \({-}{33}\) | \(221\) | \(28\) | \(400\) |
| \(0\) | \(217\) | \(28\) | \(400\) |
| \(33\) | \(238\) | \(28\) | \(400\) |
| \(67\) | \(263\) | \(26\) | \(400\) |
| \(100\) | \(277\) | \(30\) | \(400\) |
| \(133\) | \(298\) | \(32\) | \(400\) |
| \(167\) | \(316\) | \(34\) | \(400\) |
| \(\infty\) | \(348\) | \(92\) | \(400\) |
Assuming that the model is correct, the observable mean reaction times follow an approximate Normal distribution around the model prediction \(E(T_\tau)\) for each condition. We can, therefore, use a standard goodness-of-fit measure by \(z\)-standardization.
z <- function(m, s, n, tau, c, mu_A, sigma_A, mu_B, sigma_B, mu_M)
dfrac(m - denote(mu[tau], ET(tau, c, mu_A, sigma_A, mu_B, sigma_B, mu_M),
"the expected mean response time"),
s / sqrt(n))
math(call("=", quote(z[tau]), z))\({{z}_{\tau}}{=}{\displaystyle{\frac{{m}{-}{{\mu}_{\tau}}}{{s}{/}{\sqrt{n}}}}}\), with \({{\mu}_{\tau}}{=}{E{\left({{T}_{\tau}}{;}{{{\mu}_{\mathrm{A}}}{{,}{{\sigma}_{\mathrm{A}}}{{,}{{\mu}_{\mathrm{B}}}{{,}{{\sigma}_{\mathrm{B}}}{{,}{{\mu}_{\mathrm{M}}}}}}}}\right)}}\) denoting the expected mean response time
The overall goodness-of-fit is the sum of the squared \(z\)-statistics for each onset asynchrony. Assuming again that the architecture of the model is correct, but the parameters are adjusted to the data, it follows a \(\chi^2(8\ \mathrm{df})\)-distribution.
zv <- Vectorize(z, vectorize.args = c('m', 's', 'n', 'tau'))
hook(zv(.M, .S, .N, .Tau, .C, .MA, .SA, .MB, .SB, .MM), z[.Tau])
gof <- function(par, tau, m, s, n)
sum(zv(m, s, n, tau, c=100L, mu_A=par["mu_A"], sigma_A=par["sigma_A"],
mu_B=par["mu_B"], sigma_B=par["sigma_B"], mu_M=par["mu_M"])^2L)
math(call("=", quote(X["8 df"]^2L), gof))\({{X}_{\mathrm{8 df}}^{2}}{=}{\sum{{z}_{\tau}^{2}}}\)
with the degrees of freedom given by the difference between the number of observations (13) and the number of free model parameters \(\theta = \langle\mu_{\mathrm A}, \sigma_{\mathrm A}, \mu_{\mathrm B}, \sigma_{\mathrm B}, \mu_{\mathrm M}\rangle\); the barrier \(c\) is only a scaling parameter.
\({\hat{\theta}}{=}{\arg\min{gof{\left(\theta\right)}}}\)
The best fitting parameter values and their confidence intervals are given in Table 5.
The goodness-of-fit statistic indicates some lack of fit, \(X^2(8\ \mathrm{df}) = 28.34, p = 0.0004\). Given the large trial numbers in the original study, this is not an unexpected result. For more detail, especially on fitting the observed standard deviations, the reader is referred to the original paper (Schwarz 1994).
| Parameter | Estimate | CI |
|---|---|---|
| \({\mu}_{\mathrm{A}}\) | \(0.53\) | \(\left({0.51}{{,}{0.55}}\right)\) |
| \({\sigma}_{\mathrm{A}}\) | \(4.55\) | \(\left({3.95}{{,}{5.16}}\right)\) |
| \({\mu}_{\mathrm{B}}\) | \(1.36\) | \(\left({1.23}{{,}{1.49}}\right)\) |
| \({\sigma}_{\mathrm{B}}\) | \(13.46\) | \(\left({7.80}{{,}{19.11}}\right)\) |
| \({\mu}_{\mathrm{M}}\) | \(161.09\) | \(\left({156.91}{{,}{165.28}}\right)\) |
This package allows R to render its terms in pretty mathematical equations. It extends the current features of R and existing packages for displaying mathematical formulas in R (Murrell and Ihaka 2000), but most importantly, mathml bridges the gap between computational needs, presentation of results, and their reproducibility. The package supports both MathML and LaTeX/MathJax for use in R Markdown documents, presentations and Shiny App webpages.
Researchers or teachers can already use R Markdown to conduct analyses and show results, and mathml smoothes this process and allows for integrated calculations and output. As shown in the case study of the previous section, mathml can help to improve data analyses and statistical reports from an aesthetical perspective, as well as regarding reproducibility of research.
Furthermore, the package may also allow for a better detection of possible mistakes in R programs. Similar to most programming languages (Green 1977), R code is notoriously hard to read, and the poor legibility of the language is one of the main sources of mistakes. For illustration, we consider again Equation 10 in Schwarz (1994).
f1 <- function(tau)
{ dfrac(c, mu_A) + (dfrac(1L, mu_A) - dfrac(1L, mu_A + mu_B) *
((mu_A*tau - c) * pnorm(dfrac(c - mu_A*tau, sqrt(sigma_A^2L*tau)))
- (mu_A*tau + c) * exp(dfrac(2L*mu_A*tau, sigma_A^2L))
* pnorm(dfrac(-c - mu_A*tau, sqrt(sigma_A^2L*tau)))))
}
math(f1)\({\displaystyle{\frac{c}{{\mu}_{\mathrm{A}}}}}{+}{\left\{{\displaystyle{\frac{1}{{\mu}_{\mathrm{A}}}}}{-}{{\displaystyle{\frac{1}{{{\mu}_{\mathrm{A}}}{+}{{\mu}_{\mathrm{B}}}}}}{\cdot}{\left[{{\left({{{\mu}_{\mathrm{A}}}{{}}{\tau}}{-}{c}\right)}{\cdot}{\Phi{\left(\displaystyle{\frac{{c}{-}{{{\mu}_{\mathrm{A}}}{{}}{\tau}}}{\sqrt{{{\sigma}_{\mathrm{A}}^{2}}{{}}{\tau}}}}\right)}}}{-}{{{\left({{{\mu}_{\mathrm{A}}}{{}}{\tau}}{+}{c}\right)}{\cdot}{\exp{\left(\displaystyle{\frac{{{2}{{}}{{\mu}_{\mathrm{A}}}}{{}}{\tau}}{{\sigma}_{\mathrm{A}}^{2}}}\right)}}}{\cdot}{\Phi{\left(\displaystyle{\frac{{{-}{c}}{-}{{{\mu}_{\mathrm{A}}}{{}}{\tau}}}{\sqrt{{{\sigma}_{\mathrm{A}}^{2}}{{}}{\tau}}}}\right)}}}\right]}}\right\}}\)
The first version has a wrong parenthesis, which is barely visible in the code, whereas in the mathematical representation, the wrong curly brace is immediately obvious (the correct version is shown below for comparison).
f2 <- function(tau)
{ dfrac(c, mu_A) + (dfrac(1L, mu_A) - dfrac(1L, mu_A + mu_B)) *
((mu_A*tau - c) * pnorm(dfrac(c - mu_A*tau, sqrt(sigma_A^2L*tau)))
- (mu_A*tau + c) * exp(dfrac(2L*mu_A*tau, sigma_A^2L))
* pnorm(dfrac(-c - mu_A*tau, sqrt(sigma_A^2L*tau))))
}
math(f2)\({\displaystyle{\frac{c}{{\mu}_{\mathrm{A}}}}}{+}{{\left({\displaystyle{\frac{1}{{\mu}_{\mathrm{A}}}}}{-}{\displaystyle{\frac{1}{{{\mu}_{\mathrm{A}}}{+}{{\mu}_{\mathrm{B}}}}}}\right)}{\cdot}{\left[{{\left({{{\mu}_{\mathrm{A}}}{{}}{\tau}}{-}{c}\right)}{\cdot}{\Phi{\left(\displaystyle{\frac{{c}{-}{{{\mu}_{\mathrm{A}}}{{}}{\tau}}}{\sqrt{{{\sigma}_{\mathrm{A}}^{2}}{{}}{\tau}}}}\right)}}}{-}{{{\left({{{\mu}_{\mathrm{A}}}{{}}{\tau}}{+}{c}\right)}{\cdot}{\exp{\left(\displaystyle{\frac{{{2}{{}}{{\mu}_{\mathrm{A}}}}{{}}{\tau}}{{\sigma}_{\mathrm{A}}^{2}}}\right)}}}{\cdot}{\Phi{\left(\displaystyle{\frac{{{-}{c}}{-}{{{\mu}_{\mathrm{A}}}{{}}{\tau}}}{\sqrt{{{\sigma}_{\mathrm{A}}^{2}}{{}}{\tau}}}}\right)}}}\right]}}\)
As the reader may know from their own experience, missed parentheses are frequent causes of wrong results and errors that are hard to locate in programming code. This particular example shows that mathematical rendering can help to substantially reduce the amount of careless errors in programming.
One limitation of the package is the lack of a convenient way to insert line breaks. This is mostly due to lacking support by MathML and LaTeX renderers. For example, in its current stage, the LaTeX package breqn (Robertson et al. 2021) is mostly a proof of concept. Moreover, mathml only works in one direction, that is, it is not possible to translate from LaTeX or HTML back to R (see Capretto 2023, for an example).
The package mathml is available for R version 4.2 and later, and can be
easily installed using the usual install.packages("mathml"). At its
present stage, it supports output in HTML, LaTeX, and Microsoft Word
(via pandoc, MacFarlane 2022). The source code of the package is found
at https://github.com/mgondan/mathml.
For convenience, the translation of the R expressions is achieved through a Prolog interpreter provided by another R package rolog (Gondan 2022). If a version of SWI-Prolog (Wielemaker et al. 2012) is found on the system, rolog connects to it. Alternatively, the SWI-Prolog runtime libraries can be conveniently accessed by installing the R package rswipl (Gondan 2023). Prolog is a classical logic programming language with many applications in expert systems, computer linguistics and symbolic artificial intelligence. The strength of Prolog lies in its concise representation of facts and rules for knowledge and grammar, as well as its efficient built-in search engine for closed world domains. Whereas Prolog is weak in statistical computation, but strong in symbolic manipulation, the converse may be said for the R language. rolog bridges this gap by providing an interface to a SWI-Prolog distribution (Wielemaker et al. 2012) in R. The communication between the two systems is mainly in the form of queries from R to Prolog, but two Prolog functions allow ring back and evaluation of terms in R.
The proper term for a Prolog “function” is predicate, and it is
typically written with name and arity (i.e., number of arguments),
separated by a forward slash. Thus, at the Prolog end, the predicate
math/2 translates the representation of the R call pbinom(K, N, Pi)
into a more general representation of an R function fn/2 with the name
P_Bi, one argument X =< K, and the two parameters N and Pi, as
shown below.
math(pbinom(K, N, Pi), M)
=> M = fn(subscript('P', "Bi"), (['X' =< K] ; [N, Pi])).math/2 operates like a macro that translates one mathematical element
(here, pbinom(K, N, Pi)) to another mathematical element, namely
fn(Name, (Args ; Pars)). The low-level predicate ml/3 is used to
convert these basic elements to MathML.
ml(fn(Name, (Args ; Pars)), M, Flags)
=> ml(Name, N, Flags),
ml(paren(list(op(;), [list(op(','), Args), list(op(','), Pars)])), X, Flags),
M = mrow([N, mo(&(af)), X]).The relevant rule for ml/3 builds the MathML entity
mrow([N, mo(&(af)), X]), with N representing the name of the
function and X its arguments and parameters enclosed in parentheses. A
corresponding rule jax/3 does the same for MathJax/LaTeX. A list of
flags can be used for context-sensitive translation (see, e.g., the
section on errors above).
Several ways exist for translating new R terms to their mathematical representation. We have already seen above how to use “hooks” to translate long variable names from R to compact mathematical signs, as well as functions such as cumulative probabilities \(P(X \le k)\) to different representations like \(\sum_{i=0}^k P(X = i)\). Obviously, the hooks require that there already exists a rule to translate the target representation into MathML and MathJax.
In this appendix we describe a few more ways to extend the set of
translations according to a user’s needs. As stated in the background
section, the Prolog end provides two classes of rules for translation,
macros math/2,3,4 mirroring the R hooks mentioned above, and the
low-level predicates ml/3 and jax/3 that create proper MathML and
LaTeX terms.
To render the model equation of a linear model such as
lm(EOT ~ T0 + Therapy, data=d) in mathematical form (see also Anderson et al. 2023),
it is sufficient to map the
Formula in lm(Formula, Data) to its respective equation. This can be
done in two ways, using either the hooks described above, or a new
math/2 macro at the Prolog end.
hook(lm(.Formula, .Data), .Formula)The hook is simple, but is a bit limited because only R’s tilde-form of linear models is shown, and it only works for a call with exactly two arguments.
Below is an example of how to build a linear equation of the form \(Y = b_0 + b_1X_1 + ...\) using the Prolog macros from mathml.
math_hook(LM, M) :-
compound(LM),
LM =.. [lm, ~(Y, Sum) | _Tail],
summands(Sum, Predictors),
findall(subscript(b, X) * X, member(X, Predictors), Terms),
summands(Model, Terms),
M = (Y == subscript(b, 0) + Model + epsilon).The predicate summands/2 unpacks an expression A + B + C to a list
[C, B, A] and vice-versa (see the file lm.pl for details).
rolog::consult(system.file(file.path("pl", "lm.pl"), package="mathml"))
term <- quote(lm(EOT ~ T0 + Therapy, data=d, na.action=na.fail))
math(term)\({EOT}{=}{{{{b}_{0}}{+}{{{{b}_{T0}}{{}}{T0}}{+}{{{b}_{Therapy}}{{}}{Therapy}}}}{+}{\epsilon}}\)
See the section above on short mathematical names for replacing lengthy R labels.
Base R does not provide a function like cuberoot(x) or
nthroot(x, n), and the present package does not support the respective
representation. To obtain a cube root, a programmer would typically type
x^(1/3) or better x^{1/3} (see the practice section why the curly
brace is preferred in an exponent), resulting in \(x^{1/3}\) which may
still not match everyone’s taste. Here we describe the steps needed to
represent the \(n\)-th root as \(\sqrt[n]x\).
We assume that nthroot(x, n) is available in the current namespace
(manually defined, or from R package
pracma, Borchers 2022), so
that the names of the arguments and their order are accessible to
canonical() if needed. As we can see below, mathml uses a default
representation name(arguments) for such unknown functions.
nthroot <- function(x, n)
x^{1L/n}
term <- canonical(quote(nthroot(n=3L, 2L)))
math(term)\(nthroot{\left({2}{{,}{3}}\right)}\)
A proper MathML term is obtained by mlx/3 (the x in mlx indicates that
it is an extension and is prioritized over the default ml/3 rules).
mlx/3 recursively invokes ml/3 for translating the function
arguments X and N, and then constructs the correct MathML entity
<mroot>...</mroot>.
mlx(nthroot(X, N), M, Flags) :-
ml(X, X1, Flags),
ml(N, N1, Flags),
M = mroot([X1, N1]).The explicit unification M = ... in the last line serves to avoid
clutter in the head of mlx/3. The Prolog file nthroot.pl also
includes the respective rule for LaTeX and can be consulted from the
package folder via the underlying package rolog.
rolog::consult(system.file(file.path("pl", "nthroot.pl"), package="mathml"))
term <- quote(nthroot(a * (b + c), 3L)^2L)
math(term)\({\left[\sqrt[3]{{a}{\cdot}{\left({b}{+}{c}\right)}}\right]}^{2}\)
term <- quote(a^(1L/3L) + a^{1L/3L} + a^(1.0/3L))
math(term)\({{\sqrt[3]{a}}{+}{{a}^{{1}{/}{3}}}}{+}{{a}^{\left({1.00}{/}{3}\right)}}\)
The file nthroot.pl includes three more statements precx/3 and
parenx/3, as well as a math_hook/2 macro. The first sets the
operator precedence of the cubic root above the power, thereby putting a
parentheses around nthroot in \((\sqrt[3]{\ldots})^2\). The second tells
the system to increase the counter of the parentheses below the root,
such that the outer parenthesis becomes a square bracket.
The last rule maps powers like a^(1L/3L) to nthroot/3, as shown in
the first summand. Of course, mathml is not a proper computer algebra
system. As is illustrated by the other terms in the sum, such macros are
limited to purely syntactical matching, and terms like a^{1L/3L} with
the curly brace or a^(1.0/3L) with a floating point number in the
numerator are not detected.
Supported by the Erasmus+ program of the European Commission (2019-1-EE01-KA203-051708).
Supplementary materials are available in addition to this article. It can be downloaded at RJ-2023-073.zip
Text and figures are licensed under Creative Commons Attribution CC BY 4.0. The figures that have been reused from other sources don't fall under this license and can be recognized by a note in their caption: "Figure from ...".
For attribution, please cite this work as
Gondan & Alfarone, "mathml: Translate R Expressions to MathML and LaTeX", The R Journal, 2023
BibTeX citation
@article{RJ-2023-073,
author = {Gondan, Matthias and Alfarone, Irene},
title = {mathml: Translate R Expressions to MathML and LaTeX},
journal = {The R Journal},
year = {2023},
note = {https://doi.org/10.32614/RJ-2023-073},
doi = {10.32614/RJ-2023-073},
volume = {15},
issue = {3},
issn = {2073-4859},
pages = {277-291}
}