1e379e7635
Conflicts: C/c_interface.c C/exec.c |
||
---|---|---|
.. | ||
doc | ||
examples | ||
CMakeLists.txt | ||
configure.in | ||
LICENSE | ||
Makefile_so | ||
Makefile.in | ||
pack.pl | ||
pltotex.pl | ||
rconfig.h.cmake | ||
rconfig.h.in | ||
README.md | ||
real.c | ||
real.doc | ||
real.h | ||
real.pl |
The R Prolog Progrmming Interface
@file real.pl @author Nicos Angelopoulos @author Vitor Santos Costa @version 1:0:4, 2013/12/25, sinter_class @license Perl Artistic License @defgroup libReal An interface to the R statistical software. @ingroup packages
This library enables the communication with an R process started as a shared library. It is the result of the efforts of two research groups that have worked in parallel. The syntactic emphasis on a minimalistic interface.
In the doc/ directory of the distribution there is user's guide, a published paper
and html documentation from PlDoc (doc/html/real.html). There is large number
of examples in examples/for_real.pl
.
A single predicate (<-/2,<-/1) channels the bulk of the interactions. In addition to using R as a shared library, real uses the c-interfaces of SWI/Yap and R to pass objects in both directions. The usual mode of operation is to load Prolog values on to R variables and then call R functions on these values. The return value of the called function can be either placed on R variable or passed back to Prolog. It has been tested extensively on current SWI and YAP on Linux machines but it should also compile and work on MS operating systems and Macs.
The main modes for utilising the interface are
<- +Rexpr
<- +Rvar
Print Rvar or evaluate expression Rexpr in R
+Rvar <- +PLdata
+Rexpr <- +PLdata
-PLvar <- +Rvar
-PLvar <- +Rexpr
+Rexpr1 <- +Rexpr2
Pass Prolog data to R, pass R data to Prolog or assign an R expression to an assignable R expression.
@defgroup TestingR Testing Real
There is a raft of examples packed in a singl```e file that test the library.
?- [pack(real/examples/for_real)].
?- for_real.
?- edit( pack(real/examples/for_real) ).
@defgroup RSyntax Prolog and R Syntax
There are syntactic conventions in R that make unparsable prolog code. Notably function and variable names are allowed to contain dots, square brackets are used to access parts of vectors and arrays and functions are allowed empty argument tuples. We have introduced relevant syntax which allows for easy transition between prolog and R. Prolog constructs are converted by the library as follows:
- =|..|= within atoms -> =|.|= (ex. =| as..integer(c(1,2,3)) -> as.integer(c(1,2,3))|= )
- =|^[]|= after atoms -> =|[]|= (ex. =|a^[2] -> a[2] |=)
- =|(.)|= at the end of atoms that are known R functions -> =|()|= (ex. =|dev..off(.) -> dev.off()|= )
- =|[]|= -> c() (which equal to R's NULL value)
- ( f(x) :- (..)) -> f(x) (...)
- Lists of lists are converted to matrices. All first level lists must have the same length.
- Filenames must be given as Prolog strings.
- R specific operators (eg. %*% should be quoted in Prolog.
-
- prepends strings, for (Prolog) atoms: +'String'
- Expressions that pose difficulty in translation can always be passed as unquoted Prolog atoms or strings.
]]* since 0:1:2 foo() is valid syntax: =|<- dev..off() |= works now (with no need for dev..off(.))
- since 0:1:2 mat[1] is valid syntax: =|m[1] <- 4|= works now (with no need for m^[...])
@defgroup RDataTransfer Mapping Data betweenn Prolog and R
R vectors are mapped to prolog lists and matrices are mapped to nested lists. The convention works the other way around too.
There are two ways to pass prolog data to R. The more efficient one is by using
Rvar <- PLdata
Where Pldata is one of the basic data types (number,boolean) a list or a c/n term. This transfers via C data between R and Prolog. In what follows atomic PLval data are simply considered as singleton lists. Flat Pldata lists are translated to R vectors and lists of one level of nesting to R matrices (which are 2 dimensional arrays in R parlance). The type of values of the vector or matrice is taken to be the type of the first data element of the Pldata according to the following :
* integer -> integer
* float -> double
* atom -> char
* boolean -> logical
Booleans are represented in prolog as true/false atoms. Currently arrays of aribtrary dimensions are not supported in the low-level interface. Note that in R a scalar is just a one element vector. When passing non-scalars the interface will assume the type of the object is that of the first scalar until it encounters something different. Real will currently re-start and repopulate partial integers for floats as illustrated below:
r <- [1,2,3]. % pass 1,2,3 to an R vector r
R <- r. % pass contents of R vector r to Prolog variable R
R = [1, 2, 3].
i <- [1,2,3.1]. % r is now a vector of floats, rather than integers
I <- i.
I = [1.0, 2.0, 3.1].
However, not all possible "corrections" are currently supported. For instance,
?- c <- [a,b,c,1].
ERROR: real:set_R_variable/2: Type error: `boolean' expected, found `a'
In the data passing mode we map Prolog atoms to R strings-
?- x <- [abc,def].
true.
?- <- x.
[1] "abc" "def"
true.
?- X <- x.
X = [abc, def].
In addition, Prolog data can be passed through the expression mechanism. That is, data appearing in an arbitrary R expression will be parsed and be part of the long string that will be passed from Prolog to R for evaluation. This is only advisable for short data structures. For instance,
tut_4a :-
state <- c(+"tas", +"sa", +"qld", +"nsw", +"nsw"),
<- state.
tut_4b :-
state <- c(+tas, +sa, +qld, +nsw, +nsw),
<- state.
Through this interface it is more convenient to be explicit about R chars by Prolog prepending atoms or codes with + as in the above example.
@defgroup RealExamples Examples
?- e <- numeric(.).
yes
?- e^[3] <- 17.
yes
?- e[3] <- 17.
yes
?- Z <- e.
Z = ['$NaN','$NaN',17.0]
?- e^[10] <- 12.
yes
?- Z <- e.
Z = ['$NaN','$NaN',17.0,'$NaN','$NaN','$NaN','$NaN','$NaN','$NaN',12.0]
rtest :-
y <- rnorm(50), % get 50 random samples from normal distribution
<- y, % print the values via R
x <- rnorm(y), % get an equal number of normal samples
<- x11(width=5,height=3.5), % create a plotting window
<- plot(x,y) % plot the two samples
r_wait, % wait for user to hit Enter
% <- dev..off(.). % old syntax, still supported
<- dev.off(). % close the plotting window. foo() now acceptable in supported Prologs
tut6 :-
d <- outer(0:9, 0:9),
fr <- table(outer(d, d, "-")),
<- plot(as..numeric(names(fr)), fr, type="h", xlab="Determinant", ylab="Frequency").
tut4b :-
state <- [tas,sa,qld,nsw,nsw,nt,wa],
statef <- factor(state),
incmeans <- tapply( c(60, 49, 40, 61, 64, 60, 59), statef, mean ),
<- incmeans.
logical :-
t <- [1,2,3,4,5,1],
s <- t~~~~1,
<- s,
S <- s,
write( s(S) ), nl.
Info
@see http://stoics.org.uk/~nicos/sware/real @see pack(real/examples/for_real) @see pack(real/doc/real.html) @see pack(real/doc/guide.pdf) @see pack(real/doc/padl2013-real.pdf) @see http://www.r-project.org/
Also @subpaage yap-real describes the YAP specfic details in real.
*/Development of real in YAP (#yap-real)
YAP includes a development version of real, designed to experiment with the internals of the implementation of R. It includes major changes and is likely to be much less stable than the version maintained by Nicos ANgelopoulos. We refer to the version herein as 'realC' and describe the main novelties vs the version described in~\cite{}. Their major differences:
-
Most of realC is written in
C
, instead of aa a Prolog string generator. TheC
code respects the SWI-Prolog fli interface and should work both in YAP and in SWI-Prolog. -
realC uses Prolog atoms to represent real variables. R sequences of characters are represented as Prolog strings (not as lists of character codes). The atoms
true
andfalse
indicate boolean constants.By default, YAP represents sequences of codes using double quotes, and strings by back quotes. Please consult the documentation o the ISO-Prolog flag
double_quotes
if you using prefer reading double-quote strings as Prolog string. -
Free variables can be used to represent missing arguments,ie. `a[_,"G23"] would represent the column "G23".
-
All recent versions of real support the common syntax extensions for [], (), thus realC allows writing `a2 <- f().
-
YAP allows A.B to be interpreted as [A|B]. This version takes advantage of this implementation quirk, and allows one to write expressions such as
a.b[2] <- f.g()
. -
The left-hand side msy be:
- a ground unary term, assumed to be an attribute
- an index
- an R variable
- a logic variable, or other Prolog term: in this case it will be unified with the result of evaluating the right-hamd side. Yap ?- [examples/for_real]. ?- for_real.
Nicos Angelopoulos and Vitor Santos Costa December, 2012.
Updates: Nicos Angelopoulos Dec. 2013, March, 2014
Updates: Vitor Santos Costa Dec. 2015