Include random effects in starting parameters

- mean_degparms() now optionally returns starting values for fixed and
  random effects, which makes it possible to obtain acceptable fits
  also in more difficult cases (with more parameters)
- Fix the anova method, as it is currently not enough to inherit from
  lme: https://bugs.r-project.org/bugzilla/show_bug.cgi?id=17761
- Show fit information, and per default also errmin information
  in plot.nlme.mmkin()
- Examples for nlme.mmkin: Decrease tolerance and increase the number of
  iterations in the PNLS step in order to be able to fit FOMC-SFO and
  DFOP-SFO

3 files changed, 131 insertions, 79 deletions

diff --git a/R/nlme.R b/R/nlme.R
index fafaa7b6..ef93327c 100644
--- a/R/nlme.R
+++ b/R/nlme.R
@@ -8,6 +8,7 @@
 #' @param object An mmkin row object containing several fits of the same model to different datasets
 #' @import nlme
 #' @rdname nlme
+#' @seealso \code{\link{nlme.mmkin}}
 #' @examples
 #' sampling_times = c(0, 1, 3, 7, 14, 28, 60, 90, 120)
 #' m_SFO <- mkinmod(parent = mkinsub("SFO"))
@@ -47,73 +48,9 @@
 #'   start = mean_dp)
 #' summary(m_nlme)
 #' plot(augPred(m_nlme, level = 0:1), layout = c(3, 1))
+#' # augPred does not seem to work on fits with more than one state
+#' # variable
 #'
-#' \dontrun{
-#'   # Test on some real data
-#'   ds_2 <- lapply(experimental_data_for_UBA_2019[6:10],
-#'    function(x) x$data[c("name", "time", "value")])
-#'   m_sfo_sfo <- mkinmod(parent = mkinsub("SFO", "A1"),
-#'     A1 = mkinsub("SFO"), use_of_ff = "min")
-#'   m_sfo_sfo_ff <- mkinmod(parent = mkinsub("SFO", "A1"),
-#'     A1 = mkinsub("SFO"), use_of_ff = "max")
-#'   m_fomc_sfo <- mkinmod(parent = mkinsub("FOMC", "A1"),
-#'     A1 = mkinsub("SFO"))
-#'   m_dfop_sfo <- mkinmod(parent = mkinsub("DFOP", "A1"),
-#'     A1 = mkinsub("SFO"))
-#'   m_sforb_sfo <- mkinmod(parent = mkinsub("SFORB", "A1"),
-#'     A1 = mkinsub("SFO"))
-#'
-#'   f_2 <- mmkin(list("SFO-SFO" = m_sfo_sfo,
-#'    "SFO-SFO-ff" = m_sfo_sfo_ff,
-#'    "FOMC-SFO" = m_fomc_sfo,
-#'    "DFOP-SFO" = m_dfop_sfo,
-#'    "SFORB-SFO" = m_sforb_sfo),
-#'     ds_2)
-#'
-#'   grouped_data_2 <- nlme_data(f_2["SFO-SFO", ])
-#'
-#'   mean_dp_sfo_sfo <- mean_degparms(f_2["SFO-SFO", ])
-#'   mean_dp_sfo_sfo_ff <- mean_degparms(f_2["SFO-SFO-ff", ])
-#'   mean_dp_fomc_sfo <- mean_degparms(f_2["FOMC-SFO", ])
-#'   mean_dp_dfop_sfo <- mean_degparms(f_2["DFOP-SFO", ])
-#'   mean_dp_sforb_sfo <- mean_degparms(f_2["SFORB-SFO", ])
-#'
-#'   nlme_f_sfo_sfo <- nlme_function(f_2["SFO-SFO", ])
-#'   nlme_f_sfo_sfo_ff <- nlme_function(f_2["SFO-SFO-ff", ])
-#'   nlme_f_fomc_sfo <- nlme_function(f_2["FOMC-SFO", ])
-#'   assign("nlme_f_sfo_sfo", nlme_f_sfo_sfo, globalenv())
-#'   assign("nlme_f_sfo_sfo_ff", nlme_f_sfo_sfo_ff, globalenv())
-#'   assign("nlme_f_fomc_sfo", nlme_f_fomc_sfo, globalenv())
-#'
-#'   # Allowing for correlations between random effects (not shown)
-#'   # leads to non-convergence
-#'   f_nlme_sfo_sfo <- nlme(value ~ nlme_f_sfo_sfo(name, time,
-#'        parent_0, log_k_parent_sink, log_k_parent_A1, log_k_A1_sink),
-#'      data = grouped_data_2,
-#'      fixed = parent_0 + log_k_parent_sink + log_k_parent_A1 + log_k_A1_sink ~ 1,
-#'      random = pdDiag(parent_0 + log_k_parent_sink + log_k_parent_A1 + log_k_A1_sink ~ 1),
-#'      start = mean_dp_sfo_sfo)
-#'    # augPred does not see to work on this object, so no plot is shown
-#'
-#'   # The same model fitted with transformed formation fractions does not converge
-#'   f_nlme_sfo_sfo_ff <- nlme(value ~ nlme_f_sfo_sfo_ff(name, time,
-#'        parent_0, log_k_parent, log_k_A1, f_parent_ilr_1),
-#'      data = grouped_data_2,
-#'      fixed = parent_0 + log_k_parent + log_k_A1 + f_parent_ilr_1 ~ 1,
-#'      random = pdDiag(parent_0 + log_k_parent + log_k_A1 + f_parent_ilr_1 ~ 1),
-#'      start = mean_dp_sfo_sfo_ff)
-#'
-#'   f_nlme_fomc_sfo <- nlme(value ~ nlme_f_fomc_sfo(name, time,
-#'        parent_0, log_alpha, log_beta, log_k_A1, f_parent_ilr_1),
-#'      data = grouped_data_2,
-#'      fixed = parent_0 + log_alpha + log_beta + log_k_A1 + f_parent_ilr_1 ~ 1,
-#'      random = pdDiag(parent_0 + log_alpha + log_beta + log_k_A1 + f_parent_ilr_1 ~ 1),
-#'      start = mean_dp_fomc_sfo)
-#'
-#'   # DFOP-SFO and SFORB-SFO did not converge with full random effects
-#'
-#'   anova(f_nlme_fomc_sfo, f_nlme_sfo_sfo)
-#' }
 #' @return A function that can be used with nlme
 #' @export
 nlme_function <- function(object) {
@@ -185,14 +122,25 @@ nlme_function <- function(object) {
 }
 
 #' @rdname nlme
-#' @return A named vector containing mean values of the fitted degradation model parameters
+#' @return If random is FALSE (default), a named vector containing mean values
+#'   of the fitted degradation model parameters. If random is TRUE, a list with
+#'   fixed and random effects, in the format required by the start argument of
+#'   nlme for the case of a single grouping variable ds?
+#' @param random Should a list with fixed and random effects be returned?
 #' @export
-mean_degparms <- function(object) {
+mean_degparms <- function(object, random = FALSE) {
   if (nrow(object) > 1) stop("Only row objects allowed")
   degparm_mat_trans <- sapply(object, parms, transformed = TRUE)
   mean_degparm_names <- setdiff(rownames(degparm_mat_trans), names(object[[1]]$errparms))
-  res <- apply(degparm_mat_trans[mean_degparm_names, ], 1, mean)
-  return(res)
+  fixed <- apply(degparm_mat_trans[mean_degparm_names, ], 1, mean)
+  if (random) {
+    degparm_mat_trans[mean_degparm_names, ]
+    random <- t(apply(degparm_mat_trans[mean_degparm_names, ], 2, function(column) column - fixed))
+    rownames(random) <- as.character(1:nrow(random))
+    return(list(fixed = fixed, random = list(ds = random)))
+  } else {
+    return(fixed)
+  }
 }
 
 #' @rdname nlme
diff --git a/R/nlme.mmkin.R b/R/nlme.mmkin.R
index 2ee46f33..6e3467ed 100644
--- a/R/nlme.mmkin.R
+++ b/R/nlme.mmkin.R
@@ -32,9 +32,52 @@
 #' f <- mmkin("SFO", ds, quiet = TRUE, cores = 1)
 #' library(nlme)
 #' f_nlme <- nlme(f)
-#' nlme(f, random = parent_0 ~ 1)
-#' f_nlme <- nlme(f, start = c(parent_0 = 100, log_k_parent_sink = 0.1))
-#' update(f_nlme, random = parent_0 ~ 1)
+#' print(f_nlme)
+#' f_nlme_2 <- nlme(f, start = c(parent_0 = 100, log_k_parent_sink = 0.1))
+#' update(f_nlme_2, random = parent_0 ~ 1)
+#' \dontrun{
+#'   # Test on some real data
+#'   ds_2 <- lapply(experimental_data_for_UBA_2019[6:10],
+#'    function(x) x$data[c("name", "time", "value")])
+#'   m_sfo_sfo <- mkinmod(parent = mkinsub("SFO", "A1"),
+#'     A1 = mkinsub("SFO"), use_of_ff = "min", quiet = TRUE)
+#'   m_sfo_sfo_ff <- mkinmod(parent = mkinsub("SFO", "A1"),
+#'     A1 = mkinsub("SFO"), use_of_ff = "max", quiet = TRUE)
+#'   m_fomc_sfo <- mkinmod(parent = mkinsub("FOMC", "A1"),
+#'     A1 = mkinsub("SFO"), quiet = TRUE)
+#'   m_dfop_sfo <- mkinmod(parent = mkinsub("DFOP", "A1"),
+#'     A1 = mkinsub("SFO"), quiet = TRUE)
+#'
+#'   f_2 <- mmkin(list("SFO-SFO" = m_sfo_sfo,
+#'    "SFO-SFO-ff" = m_sfo_sfo_ff,
+#'    "FOMC-SFO" = m_fomc_sfo,
+#'    "DFOP-SFO" = m_dfop_sfo),
+#'     ds_2, quiet = TRUE)
+#'   plot(f_2["SFO-SFO", 3:4]) # Separate fits for datasets 3 and 4
+#'
+#'   f_nlme_sfo_sfo <- nlme(f_2["SFO-SFO", ])
+#'   # plot(f_nlme_sfo_sfo) # not feasible with pkgdown figures
+#'   plot(f_nlme_sfo_sfo, 3:4) # Global mixed model: Fits for datasets 3 and 4
+#'
+#'   # With formation fractions
+#'   f_nlme_sfo_sfo_ff <- nlme(f_2["SFO-SFO-ff", ])
+#'   plot(f_nlme_sfo_sfo_ff, 3:4) # chi2 different due to different df attribution
+#'
+#'   # For more parameters, we need to increase pnlsMaxIter and the tolerance
+#'   # to get convergence
+#'   f_nlme_fomc_sfo <- nlme(f_2["FOMC-SFO", ],
+#'     control = list(pnlsMaxIter = 100, tolerance = 1e-4), verbose = TRUE)
+#'   f_nlme_dfop_sfo <- nlme(f_2["DFOP-SFO", ],
+#'     control = list(pnlsMaxIter = 120, tolerance = 5e-4), verbose = TRUE)
+#'   plot(f_2["FOMC-SFO", 3:4])
+#'   plot(f_nlme_fomc_sfo, 3:4)
+#'
+#'   plot(f_2["DFOP-SFO", 3:4])
+#'   plot(f_nlme_dfop_sfo, 3:4)
+#'
+#'   anova(f_nlme_dfop_sfo, f_nlme_fomc_sfo, f_nlme_sfo_sfo)
+#'   anova(f_nlme_dfop_sfo, f_nlme_sfo_sfo) # if we ignore FOMC
+#' }
 # Code inspired by nlme.nlsList
 nlme.mmkin <- function(model, data = sys.frame(sys.parent()),
   fixed, random = fixed,
@@ -68,7 +111,7 @@ nlme.mmkin <- function(model, data = sys.frame(sys.parent()),
   thisCall[["data"]] <- nlme_data(model)
 
   if (missing(start)) {
-    thisCall[["start"]] <- mean_dp
+    thisCall[["start"]] <- mean_degparms(model, random = TRUE)
   }
 
   thisCall[["fixed"]] <- lapply(as.list(dp_names), function(el)
@@ -84,3 +127,52 @@ nlme.mmkin <- function(model, data = sys.frame(sys.parent()),
   return(val)
 }
 
+#' @export
+#' @rdname nlme.mmkin
+#' @param x An nlme.mmkin object to print
+#' @param data Should the data be printed?
+#' @param ... Further arguments as in the generic
+print.nlme.mmkin <- function(x, ...) {
+  x$call$data <- "Not shown"
+  NextMethod("print", x)
+}
+
+#' @export
+#' @rdname nlme.mmkin
+#' @param object An nlme.mmkin object to update
+#' @param ... Update specifications passed to update.nlme
+update.nlme.mmkin <- function(object, ...) {
+  res <- NextMethod()
+  res$mmkin_orig <- object$mmkin_orig
+  class(res) <- c("nlme.mmkin", "nlme", "lme")
+  return(res)
+}
+
+# The following is necessary as long as R bug 17761 is not fixed
+# https://bugs.r-project.org/bugzilla/show_bug.cgi?id=17761
+#' @export
+anova.nlme.mmkin <- function(object, ...) {
+  thisCall <- as.list(match.call())[-1]
+  object_name <- as.character(thisCall[[1]])
+  other_object_names <- sapply(thisCall[-1], as.character)
+
+  remove_class <- function(object, classname) {
+    old_class <- class(object)
+    class(object) <- setdiff(old_class, classname)
+    return(object)
+  }
+  object <- remove_class(object, "nlme.mmkin")
+  other_objects <- list(...)
+  other_objects <- lapply(other_objects, remove_class, "nlme.mmkin")
+
+  env <- new.env()
+  assign(object_name, object, env)
+  for (i in seq_along(other_objects)) {
+    assign(other_object_names[i], other_objects[[i]], env)
+  }
+  res <- eval(parse(text = paste0("anova.lme(", object_name, ", ",
+        paste(other_object_names, collapse = ", "), ")")),
+    envir = env)
+
+  return(res)
+}
diff --git a/R/plot.nlme.mmkin.R b/R/plot.nlme.mmkin.R
index ef6d100a..0f3ad715 100644
--- a/R/plot.nlme.mmkin.R
+++ b/R/plot.nlme.mmkin.R
@@ -11,6 +11,12 @@
 #' @param standardized Should the residuals be standardized? This option
 #'   is passed to \code{\link{mkinresplot}}, it only takes effect if
 #'   `resplot = "time"`.
+#' @param show_errmin Should the chi2 error level be shown on top of the plots
+#'   to the left?
+#' @param errmin_var The variable for which the FOCUS chi2 error value should
+#'   be shown.
+#' @param errmin_digits The number of significant digits for rounding the FOCUS
+#'   chi2 error percentage.
 #' @param cex Passed to the plot functions and \code{\link{mtext}}.
 #' @param rel.height.middle The relative height of the middle plot, if more
 #'   than two rows of plots are shown.
@@ -25,16 +31,19 @@
 #'  function(x) subset(x$data[c("name", "time", "value")], name == "parent"))
 #' f <- mmkin("SFO", ds, quiet = TRUE, cores = 1)
 #' #plot(f) # too many panels for pkgdown
+#' plot(f[, 3:4])
 #' library(nlme)
 #' f_nlme <- nlme(f)
 #'
 #' #plot(f_nlme) # too many panels for pkgdown
-#' plot(f_nlme, 1:2)
+#' plot(f_nlme, 3:4)
 #' @export
 plot.nlme.mmkin <- function(x, i = 1:ncol(x$mmkin_orig),
   main = "auto", legends = 1,
   resplot = c("time", "errmod"),
   standardized = FALSE,
+  show_errmin = TRUE,
+  errmin_var = "All data", errmin_digits = 3,
   cex = 0.7, rel.height.middle = 0.9,
   ymax = "auto", ...)
 {
@@ -79,13 +88,14 @@ plot.nlme.mmkin <- function(x, i = 1:ncol(x$mmkin_orig),
     state_ini[names(odeini_optim)] <- odeini_optim
 
     odeparms <- fit_a$bparms.ode
-    odeparms[names(odeparms)] <- odeparms_optim
+    odeparms[names(odeparms_optim)] <- odeparms_optim
 
     mkinfit_call[["observed"]] <- ds[[a]]
     mkinfit_call[["parms.ini"]] <- odeparms
     mkinfit_call[["state.ini"]] <- state_ini
 
-    mkinfit_call[["control"]] <- list(iter.max = 1)
+    mkinfit_call[["control"]] <- list(iter.max = 0)
+    mkinfit_call[["quiet"]] <- TRUE
 
     res <- suppressWarnings(do.call("mkinfit", mkinfit_call))
     return(res)
@@ -94,9 +104,11 @@ plot.nlme.mmkin <- function(x, i = 1:ncol(x$mmkin_orig),
   # Set dimensions with names and the class (mmkin)
   attributes(mmkin_nlme) <- attributes(x$mmkin_orig[, i])
 
-  plot(mmkin_nlme[, i], main = main, legends = legends,
+  plot(mmkin_nlme, main = main, legends = legends,
     resplot = resplot, standardized = standardized,
-    show_errmin = FALSE, cex = cex,
+    show_errmin = show_errmin,
+    errmin_var = errmin_var, errmin_digits = errmin_digits,
+    cex = cex,
     rel.height.middle = rel.height.middle,
     ymax = ymax, ...)