Remove treeFromContexts function and add activateFromContexts method

NAMESPACE

@@ -9,7 +9,6 @@ export(baConTree)
 export(fit_vlmc)
 export(metropolis_vlmc)
 export(rvlmc)
-export(treeFromContexts)
 importFrom(Brobdingnag,as.brob)
 importFrom(Brobdingnag,cbrob)
 importFrom(R6,R6Class)

NEWS.md

@@ -2,6 +2,12 @@
 
 ## New features
 
+* `treeFromContexts()` has been removed. Instead, `ContextTree` now exposes an
+  `$activateFromContexts(contexts)` method that activates a specific tree on an
+  existing `ContextTree` object. It accepts the same input format as the old
+  function (a character vector of context paths, or a single brace-enclosed
+  comma-separated string).
+
 * `baConTree` nodes now store `priorBranchingProbability` and
   `posteriorBranchingProbability` in their `extra` list. These are computed
   during initialisation as `prod(children sigmaPrior) / sigmaPrior` (and the

R/ContextTree.R

@@ -23,6 +23,7 @@
 #'
 #' @importFrom purrr map_chr map_lgl map map_int
 #' @importFrom glue glue
+#' @importFrom stringr str_split str_replace_all str_count
 #' @export
 ContextTree <- R6Class(
   "ContextTree",
@@ -178,6 +179,50 @@ ContextTree <- R6Class(
         })])
     },
 
+    #' @description
+    #' Sets the active tree to match a specified set of contexts, given as a
+    #' character vector or a brace-enclosed comma-separated string. The contexts
+    #' must be compatible with the tree's existing alphabet and maximal depth.
+    #' @param contexts Character vector or string. A vector of context paths
+    #'   (e.g., \code{c("*.a", "*.b.a", "*.b.b")}) or a single brace-enclosed
+    #'   string (e.g., \code{"\{*.a, *.b.a, *.b.b\}"}).
+    activateFromContexts = function(contexts) {
+      if (length(contexts) == 1) {
+        raw <- str_replace_all(contexts, "\\{|\\}", "")
+        parts <- str_split(string = raw, pattern = ",")[[1]]
+        contexts <- trimws(parts)
+        contexts <- contexts[nzchar(contexts)]
+      }
+
+      if (!validate_tree_string(contexts)) {
+        stop("Contexts do not correspond to a full tree.")
+      }
+
+      context_alphabet <- setdiff(unique(unlist(str_split(contexts, "\\."))), "*")
+      if (!all(context_alphabet %in% private$Alphabet$symbols)) {
+        stop("Contexts use symbols not in the tree's alphabet.")
+      }
+
+      max_depth <- max(str_count(contexts, "\\."))
+      if (max_depth > private$maximalDepth) {
+        stop("Contexts require a greater depth than the tree's maximal depth.")
+      }
+
+      self$activateRoot()
+      current_depth <- 0
+      while (length(setdiff(self$getActiveNodes(idx = TRUE), contexts)) > 0) {
+        subcontexts <- substr(contexts, start = 1, stop = current_depth * 2 + 1)
+        for (sc in unique(subcontexts)) {
+          if (!sc %in% contexts) {
+            self$growActive(sc)
+          }
+        }
+        current_depth <- current_depth + 1
+      }
+
+      invisible(self)
+    },
+
     #' @return Returns the leaf nodes of the maximal Context Tree (regardless of
     #' the current active tree).
     getLeaves = function(idx = TRUE) {

README.Rmd

@@ -30,8 +30,7 @@ the order of dependence can vary with the observed past. `bacontrees` provides:
 - A **Bayesian** method: `metropolis_vlmc()` (and the `baConTree` class) runs a
   Metropolis-Hastings MCMC sampler to obtain a full posterior distribution over
   context trees.
-- Simulation utilities (`rvlmc()`) and construction helpers
-  (`treeFromContexts()`).
+- Simulation utilities (`rvlmc()`).
 - R6 classes (`ContextTree`, `baConTree`) that expose the full tree structure
   for building custom algorithms.
 
@@ -101,15 +100,17 @@ head(seq1, 20)
 
 ### Building a context tree manually
 
-`treeFromContexts()` creates a `ContextTree` whose active nodes exactly match a
-supplied set of contexts.
+`ContextTree$activateFromContexts()` sets the active nodes of an existing tree
+to match a supplied set of contexts.
 
-```{r treeFromContexts}
-ct <- treeFromContexts(c("*.a", "*.b", "*.c.a", "*.c.b", "*.c.c"))
-ct$getActiveNodes()
+```{r activateFromContexts}
+tree <- ContextTree$new(alphabet = c("a", "b", "c"), maximalDepth = 2)
+tree$activateFromContexts(c("*.a", "*.b", "*.c.a", "*.c.b", "*.c.c"))
+tree$getActiveNodes()
 ```
 
-You can also create a `ContextTree` directly and manipulate it:
+You can also use `$activateMaximal()` or `$activateRoot()` to switch between
+the deepest and shallowest trees:
 
 ```{r ContextTree}
 tree <- ContextTree$new(abc_list, maximalDepth = 3)
@@ -151,7 +152,7 @@ with_progress({
     n_steps       = 2000,
     max_depth     = 3,
     alpha         = 0.01,
-    context_weights = function(node) -node$getDepth() / 3,
+    context_weights = function(node) exp(-node$getDepth() / 3),
     burnin        = 200
   )
 })
@@ -174,7 +175,7 @@ For more control, use the `baConTree` R6 class directly.
 
 ```{r baConTree}
 bt <- baConTree$new(abc_list, maximalDepth = 3, alpha = 0.01,
-                    priorWeights = function(node) -node$getDepth() / 3)
+                    priorWeights = function(node) exp(-node$getDepth() / 3))
 
 with_progress(bt$runMetropolisHastings(2000))
 
@@ -198,6 +199,7 @@ The core data structure. Manages the maximal suffix tree, tracks which nodes are
 | `$activateRoot()` | Set active tree to root-only |
 | `$activateMaximal()` | Set active tree to all maximal leaves |
 | `$activateByCode(code)` | Restore a previously stored tree code |
+| `$activateFromContexts(contexts)` | Set active tree to match a context vector or string |
 | `$growActive(path)` | Grow the active tree at a given node |
 | `$pruneActive(path)` | Prune the active tree at a given node |
 | `$setData(dataset)` | Attach sequence data and compute counts |
@@ -213,7 +215,10 @@ Extends `ContextTree` with Bayesian machinery.
 
 | Method | Description |
 |--------|-------------|
-| `$new(data, maximalDepth, alpha, priorWeights)` | Construct a Bayesian context tree |
+| `$new(data, maximalDepth, alpha, priorWeights, initialTree)` | Construct a Bayesian context tree |
 | `$runMetropolisHastings(steps)` | Run the MCMC sampler |
 | `$getChain()` | Retrieve the sampled chain as a data frame |
+| `$activateMap()` | Set active tree to the MAP tree |
+| `$sampleTree(type)` | Sample a tree from the prior or posterior |
+| `$getMarginalLikelihood(log)` | Return the marginal likelihood of the data |
 

README.md

@@ -21,8 +21,7 @@ past. `bacontrees` provides:
 - A **Bayesian** method: `metropolis_vlmc()` (and the `baConTree` class)
   runs a Metropolis-Hastings MCMC sampler to obtain a full posterior
   distribution over context trees.
-- Simulation utilities (`rvlmc()`) and construction helpers
-  (`treeFromContexts()`).
+- Simulation utilities (`rvlmc()`).
 - R6 classes (`ContextTree`, `baConTree`) that expose the full tree
   structure for building custom algorithms.
 
@@ -98,16 +97,18 @@ head(seq1, 20)
 
 ### Building a context tree manually
 
-`treeFromContexts()` creates a `ContextTree` whose active nodes exactly
-match a supplied set of contexts.
+`ContextTree$activateFromContexts()` sets the active nodes of an
+existing tree to match a supplied set of contexts.
 
 ``` r
-ct <- treeFromContexts(c("*.a", "*.b", "*.c.a", "*.c.b", "*.c.c"))
-ct$getActiveNodes()
+tree <- ContextTree$new(alphabet = c("a", "b", "c"), maximalDepth = 2)
+tree$activateFromContexts(c("*.a", "*.b", "*.c.a", "*.c.b", "*.c.c"))
+tree$getActiveNodes()
 #> [1] "*.c.b" "*.c.c" "*.a"   "*.b"   "*.c.a"
 ```
 
-You can also create a `ContextTree` directly and manipulate it:
+You can also use `$activateMaximal()` or `$activateRoot()` to switch
+between the deepest and shallowest trees:
 
 ``` r
 tree <- ContextTree$new(abc_list, maximalDepth = 3)
@@ -158,7 +159,7 @@ with_progress({
     n_steps       = 2000,
     max_depth     = 3,
     alpha         = 0.01,
-    context_weights = function(node) -node$getDepth() / 3,
+    context_weights = function(node) exp(-node$getDepth() / 3),
     burnin        = 200
   )
 })
@@ -187,7 +188,7 @@ For more control, use the `baConTree` R6 class directly.
 
 ``` r
 bt <- baConTree$new(abc_list, maximalDepth = 3, alpha = 0.01,
-                    priorWeights = function(node) -node$getDepth() / 3)
+                    priorWeights = function(node) exp(-node$getDepth() / 3))
 
 with_progress(bt$runMetropolisHastings(2000))
 
@@ -219,6 +220,7 @@ attachment.
 | `$activateRoot()` | Set active tree to root-only |
 | `$activateMaximal()` | Set active tree to all maximal leaves |
 | `$activateByCode(code)` | Restore a previously stored tree code |
+| `$activateFromContexts(contexts)` | Set active tree to match a context vector or string |
 | `$growActive(path)` | Grow the active tree at a given node |
 | `$pruneActive(path)` | Prune the active tree at a given node |
 | `$setData(dataset)` | Attach sequence data and compute counts |
@@ -234,6 +236,9 @@ Extends `ContextTree` with Bayesian machinery.
 
 | Method | Description |
 |----|----|
-| `$new(data, maximalDepth, alpha, priorWeights)` | Construct a Bayesian context tree |
+| `$new(data, maximalDepth, alpha, priorWeights, initialTree)` | Construct a Bayesian context tree |
 | `$runMetropolisHastings(steps)` | Run the MCMC sampler |
 | `$getChain()` | Retrieve the sampled chain as a data frame |
+| `$activateMap()` | Set active tree to the MAP tree |
+| `$sampleTree(type)` | Sample a tree from the prior or posterior |
+| `$getMarginalLikelihood(log)` | Return the marginal likelihood of the data |