AD support. roxygen. refactored dNmixture.

DESCRIPTION

@@ -27,8 +27,9 @@ Collate:
   dHMM.R
   dOcc.R
   dNmixture.R
+  dNmixtureAD.R
   zzz.R
-RoxygenNote: 7.2.0
+RoxygenNote: 7.3.1
 Suggests: 
     rmarkdown,
     knitr,

NAMESPACE

@@ -1,4 +1,7 @@
+# Generated by roxygen2: do not edit by hand
 
+export(dBetaBinom)
+export(dBetaBinom_One)
 export(dCJS_ss)
 export(dCJS_sv)
 export(dCJS_vs)
@@ -19,19 +22,38 @@ export(dDynOcc_vvs)
 export(dDynOcc_vvv)
 export(dHMM)
 export(dHMMo)
-export(dNmixture_s)
-export(dNmixture_v)
-export(dNmixture_BNB_oneObs)
-export(dNmixture_BNB_v)
-export(dNmixture_BNB_s)
-export(dNmixture_BBP_oneObs)
-export(dNmixture_BBP_v)
-export(dNmixture_BBP_s)
+export(dNmixtureAD_BBNB_oneObs)
+export(dNmixtureAD_BBNB_s)
+export(dNmixtureAD_BBNB_v)
+export(dNmixtureAD_BBP_oneObs)
+export(dNmixtureAD_BBP_s)
+export(dNmixtureAD_BBP_v)
+export(dNmixtureAD_BNB_oneObs)
+export(dNmixtureAD_BNB_s)
+export(dNmixtureAD_BNB_v)
+export(dNmixtureAD_s)
+export(dNmixtureAD_v)
 export(dNmixture_BBNB_oneObs)
-export(dNmixture_BBNB_v)
 export(dNmixture_BBNB_s)
+export(dNmixture_BBNB_steps)
+export(dNmixture_BBNB_v)
+export(dNmixture_BBP_oneObs)
+export(dNmixture_BBP_s)
+export(dNmixture_BBP_steps)
+export(dNmixture_BBP_v)
+export(dNmixture_BNB_oneObs)
+export(dNmixture_BNB_s)
+export(dNmixture_BNB_steps)
+export(dNmixture_BNB_v)
+export(dNmixture_s)
+export(dNmixture_steps)
+export(dNmixture_v)
 export(dOcc_s)
 export(dOcc_v)
+export(nimBetaFun)
+export(nimNmixPois_logFac)
+export(rBetaBinom)
+export(rBetaBinom_One)
 export(rCJS_ss)
 export(rCJS_sv)
 export(rCJS_vs)
@@ -52,27 +74,38 @@ export(rDynOcc_vvs)
 export(rDynOcc_vvv)
 export(rHMM)
 export(rHMMo)
-export(rNmixture_s)
-export(rNmixture_v)
-export(rNmixture_BNB_oneObs)
-export(rNmixture_BNB_v)
-export(rNmixture_BNB_s)
-export(rNmixture_BBP_oneObs)
-export(rNmixture_BBP_v)
-export(rNmixture_BBP_s)
+export(rNmixtureAD_BBNB_oneObs)
+export(rNmixtureAD_BBNB_s)
+export(rNmixtureAD_BBNB_v)
+export(rNmixtureAD_BBP_oneObs)
+export(rNmixtureAD_BBP_s)
+export(rNmixtureAD_BBP_v)
+export(rNmixtureAD_BNB_oneObs)
+export(rNmixtureAD_BNB_s)
+export(rNmixtureAD_BNB_v)
+export(rNmixtureAD_s)
+export(rNmixtureAD_v)
 export(rNmixture_BBNB_oneObs)
-export(rNmixture_BBNB_v)
 export(rNmixture_BBNB_s)
-export(nimNmixPois_logFac)
+export(rNmixture_BBNB_v)
+export(rNmixture_BBP_oneObs)
+export(rNmixture_BBP_s)
+export(rNmixture_BBP_v)
+export(rNmixture_BNB_oneObs)
+export(rNmixture_BNB_s)
+export(rNmixture_BNB_v)
+export(rNmixture_s)
+export(rNmixture_v)
 export(rOcc_s)
 export(rOcc_v)
-export(nimBetaFun)
-export(dBetaBinom)
-export(dBetaBinom_One)
-export(rBetaBinom)
-export(rBetaBinom_One)
 import(nimble)
 importFrom(stats,dbinom)
+importFrom(stats,dnbinom)
+importFrom(stats,dpois)
+importFrom(stats,qnbinom)
+importFrom(stats,qpois)
+importFrom(stats,rbeta)
 importFrom(stats,rbinom)
+importFrom(stats,rnbinom)
+importFrom(stats,rpois)
 importFrom(stats,runif)
-importFrom("stats", "dpois", "qpois", "rpois", "dnbinom", "rnbinom", "rbeta", "qnbinom")

R/dBetaBinom.R

@@ -63,10 +63,11 @@ nimBetaFun <- nimbleFunction(
   run = function(a = double(0),
                  b = double(0),
                  log = logical(0)) {
-    if (log) return(lgamma(a) + lgamma(b) - lgamma(a + b))
-    else return(exp(lgamma(a) + lgamma(b) - lgamma(a + b)))
+    log_ans <- lgamma(a) + lgamma(b) - lgamma(a + b)
+    if (log) return(log_ans)
+    else return(exp(log_ans))
     returnType(double(0))
-  })
+  }, buildDerivs=list(run=list()))
 
 #' @rdname dBetaBinom
 #' @export
@@ -77,17 +78,19 @@ dBetaBinom <- nimbleFunction(
                  shape2 = double(1),
                  log = integer(0, default = 0)) {
     logprob <- 0
+    lgNp1 <- lgamma(N+1)
     for (i in 1:length(x)) {
       logprob <- logprob +
         nimBetaFun(a = x[i] + shape1[i], b = N - x[i] + shape2[i], log = TRUE) -
         nimBetaFun(a = shape1[i], b = shape2[ i], log = TRUE) +
-        lgamma(N+1) - (lgamma(x[i] + 1) + lgamma(N - x[i] + 1))
+        lgNp1 - (lgamma(x[i] + 1) + lgamma(N - x[i] + 1))
     }
 
     if (log) return(logprob)
     return(exp(logprob))
     returnType(double(0))
-  }
+  },
+  buildDerivs = list(run = list(ignore = 'i'))
 )
 
 #' @rdname dBetaBinom
@@ -107,12 +110,14 @@ dBetaBinom_One <- nimbleFunction(
     if (log) return(logprob)
     return(exp(logprob))
     returnType(double(0))
-  }
+  },
+  buildDerivs = list(run=list())
 )
 
 
 #' @rdname dBetaBinom
 #' @export
+#' @importFrom stats rbeta
 rBetaBinom <- nimbleFunction(
   run = function(n = double(0),
                  N = double(0),
@@ -129,6 +134,7 @@ rBetaBinom <- nimbleFunction(
 
 #' @rdname dBetaBinom
 #' @export
+#' @importFrom stats rbeta
 rBetaBinom_One <- nimbleFunction(
   run = function(n = double(0),
                  N = double(0),

R/dCJS.R

@@ -92,6 +92,19 @@
 #'
 #' and so on for each combination of time-dependent and time-independent parameters.
 #'
+#' @section Notes for use with automatic differentiation:
+#'
+#' The \code{dCJS_**} distributions should all work for models and algorithms
+#' that use nimble's automatic differentiation (AD) system. In that system,
+#' some kinds of values are "baked in" (cannot be changed) to the AD calculations
+#' from the first call, unless and until the AD calculations are reset. For
+#' the \code{dCJS_**} distributions, the lengths of vector inputs and the data
+#' (\code{x}) values themselves are baked in. These can be different for different
+#' iterations through a for loop (or nimble model declarations with different indices,
+#' for example), but the lengths and data values for each specific iteration
+#' will be "baked in" after the first call. \bold{In other words, it is assumed that
+#' \code{x} are data and are not going to change.}
+#'
 #' @return
 #'
 #' For \code{dCJS_**}: the probability (or likelihood) or log probability of observation vector \code{x}.

R/dDHMM.R

@@ -109,6 +109,17 @@
 #' \code{observedStates[1:T] ~ dDHMMo(initStates[1:S], observationProbs[1:S,
 #' 1:O, 1:T], transitionProbs[1:S, 1:S, 1:(T-1)], 1, T)}
 #'
+#' The \code{dDHMM[o]} distributions should work for models and algorithms that
+#' use nimble's automatic differentiation (AD) system. In that system, some
+#' kinds of values are "baked in" (cannot be changed) to the AD calculations
+#' from the first call, unless and until the AD calculations are reset. For the
+#' \code{dDHMM[o]} distributions, the sizes of the inputs and the data (\code{x})
+#' values themselves are baked in. These can be different for different
+#' iterations through a for loop (or nimble model declarations with different
+#' indices, for example), but the sizes and data values for each specific
+#' iteration will be "baked in" after the first call. \bold{In other words, it
+#' is assumed that \code{x} are data and are not going to change.}
+#'
 #' @return For \code{dDHMM} and \code{dDHMMo}: the probability (or likelihood)
 #' or log probability of observation vector \code{x}. For \code{rDHMM} and
 #' \code{rDHMMo}: a simulated detection history, \code{x}.

R/dDynOcc.R

@@ -122,6 +122,20 @@
 #' probPersist = persistence_prob,
 #' probColonize = colonization_prob[1:(T-1)], p = p[1:T, 1:O])}
 #'
+#' @section Notes for use with automatic differentiation:
+#'
+#' The \code{dDynOcc_***} distributions should all work for models and
+#' algorithms that use nimble's automatic differentiation (AD) system. In that
+#' system, some kinds of values are "baked in" (cannot be changed) to the AD
+#' calculations from the first call, unless and until the AD calculations are
+#' reset. For the \code{dDynOcc_***} distributions, the lengths or dimensions of
+#' vector and/or matrix inputs and the \code{start} and \code{end} values
+#' themselves are baked in. These can be different for different iterations
+#' through a for loop (or nimble model declarations with different indices, for
+#' example), but the for each specific iteration will be "baked in" after the
+#' first call. \bold{It is safest if one can assume that \code{x} are data and
+#' are not going to change.}
+#'
 #' @return
 #' For \code{dDynOcc_***}: the probability (or likelihood) or log probability
 #' of observation vector \code{x}.

R/dHMM.R

@@ -109,6 +109,19 @@
 #'   \code{observedStates[1:T] ~ dHMMo(initStates[1:S], observationProbs[1:S,
 #'   1:O, 1:T], transitionProbs[1:S, 1:S], 1, T)}
 #'
+#' @section Notes for use with automatic differentiation:
+#'
+#' The \code{dHMM[o]} distributions should work for models and algorithms that
+#' use nimble's automatic differentiation (AD) system. In that system, some
+#' kinds of values are "baked in" (cannot be changed) to the AD calculations
+#' from the first call, unless and until the AD calculations are reset. For the
+#' \code{dHMM[o]} distributions, the sizes of the inputs and the data (\code{x})
+#' values themselves are baked in. These can be different for different
+#' iterations through a for loop (or nimble model declarations with different
+#' indices, for example), but the sizes and data values for each specific
+#' iteration will be "baked in" after the first call. \bold{In other words, it
+#' is assumed that \code{x} are data and are not going to change.}
+#'
 #' @return For \code{dHMM} and \code{dHMMo}: the probability (or likelihood) or
 #'   log probability of observation vector \code{x}.
 #'