gnu: Add r-lbfgs.

* gnu/packages/cran.scm (r-lbfgs): New variable.

Change-Id: Ibcb6730d72419673214c35c3d025b7934653f5b4

gnu/packages/cran.scm

@@ -10273,6 +10273,34 @@ well as additional utilities such as panel and axis annotation functions.")
 evaluation (NSE) in R.")
     (license license:gpl3+)))
 
+(define-public r-lbfgs
+  (package
+    (name "r-lbfgs")
+    (version "1.2.1.2")
+    (source
+     (origin
+       (method url-fetch)
+       (uri (cran-uri "lbfgs" version))
+       (sha256
+        (base32 "04fx9v1dn8rn09h98zxp71zcxidjmhmcv190h7mvbj3pqf3dh2bv"))))
+    (properties `((upstream-name . "lbfgs")))
+    (build-system r-build-system)
+    (propagated-inputs (list r-rcpp))
+    (home-page "https://cran.r-project.org/package=lbfgs")
+    (synopsis "Limited-memory BFGS optimization")
+    (description
+     "This package provides a wrapper built around the @code{libLBFGS}
+optimization library by Naoaki Okazaki.  The lbfgs package implements both the
+Limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) and the Orthant-Wise
+Quasi-Newton Limited-Memory (OWL-QN) optimization algorithms.  The L-BFGS
+algorithm solves the problem of minimizing an objective, given its gradient,
+by iteratively computing approximations of the inverse Hessian matrix.  The
+OWL-QN algorithm finds the optimum of an objective plus the L1-norm of the
+problem's parameters.  The package offers a fast and memory-efficient
+implementation of these optimization routines, which is particularly suited
+for high-dimensional problems.")
+    (license license:gpl2+)))
+
 (define-public r-lbfgsb3c
   (package
     (name "r-lbfgsb3c")