feat: add concept of native forecast learners

mlr-org · Jan 6, 2025 · 1fb9125 · 1fb9125
1 parent 45bb4cc
commit 1fb9125
Show file tree

Hide file tree

Showing 11 changed files with 348 additions and 156 deletions.
diff --git a/DESCRIPTION b/DESCRIPTION
@@ -31,11 +31,12 @@ RoxygenNote: 7.3.2
 Collate:
     'ForecastLearner.R'
     'zzz.R'
-    'LearnerARIMA.R'
-    'LearnerFcst.R'
+    'LearnerRegrARIMA.R'
+    'LearnerRegrAutoARIMA.R'
     'MeasureDirectional.R'
     'ResamplingForecastCV.R'
     'ResamplingForecastHoldout.R'
     'TaskFcstAirpassengers.R'
     'as_task_fcst.R'
     'bibentries.R'
+    'utils.R'
diff --git a/NAMESPACE b/NAMESPACE
@@ -5,6 +5,7 @@ S3method(as_task_fcst,TaskFcst)
 S3method(as_task_fcst,data.frame)
 export(ForecastLearner)
 export(LearnerFcstARIMA)
+export(LearnerFcstAutoARIMA)
 export(ResamplingForecastCV)
 export(ResamplingForecastHoldout)
 export(as_task_fcst)

diff --git a/R/LearnerFcst.R b/R/LearnerFcst.R
diff --git a/R/LearnerARIMA.R → R/LearnerRegrARIMA.R b/R/LearnerARIMA.R → R/LearnerRegrARIMA.R
@@ -19,10 +19,17 @@ LearnerFcstARIMA = R6Class("LearnerFcstARIMA",
     #' @description
     #' Creates a new instance of this [R6][R6::R6Class] class.
     initialize = function() {
-
-      ps = ps(
-        order = p_uty(default = c(0, 0, 0), tags = "train"),
-        seasonal = p_uty(default = c(0, 0, 0), tags = "train"),
+      param_set = ps(
+        order = p_uty(
+          default = c(0L, 0L, 0L),
+          tags = "train",
+          custom_check = crate(function(x) check_integerish(x, lower = 0L, len = 3L))
+        ),
+        seasonal = p_uty(
+          default = c(0L, 0L, 0L),
+          tags = "train",
+          custom_check = crate(function(x) check_integerish(x, lower = 0L, len = 3L))
+        ),
         include.mean = p_lgl(default = TRUE, tags = "train"),
         include.drift = p_lgl(default = FALSE, tags = "train"),
         biasadj = p_lgl(default = FALSE, tags = "train"),
@@ -31,11 +38,11 @@ LearnerFcstARIMA = R6Class("LearnerFcstARIMA",
 
       super$initialize(
         id = "fcst.arima",
-        param_set = ps,
+        param_set = param_set,
         feature_types = c("logical", "integer", "numeric"),
-        packages = c("mlr3learners", "forecast"),
+        packages = c("mlr3forecast", "forecast"),
         label = "ARIMA",
-        man = "mlr3learners::mlr_learners_arima.arima"
+        man = "mlr3forecast::mlr_learners_fcst.arima"
       )
     }
   ),
@@ -44,41 +51,42 @@ LearnerFcstARIMA = R6Class("LearnerFcstARIMA",
     .max_index = NULL,
 
     .train = function(task) {
-      if (length(task$col_roles$order) == 0L) {
+      if ("ordered" %nin% task$properties) {
         stopf("%s learner requires an ordered task.", self$id)
       }
       private$.max_index = max(task$data(cols = task$col_roles$order)[[1L]])
       pv = self$param_set$get_values(tags = "train")
       if ("weights" %in% task$properties) {
         pv = insert_named(pv, list(weights = task$weights$weight))
       }
-      if (length(task$feature_names) > 0) {
-        xreg = as.matrix(task$data(cols = task$feature_names))
+
+      if (is_task_featureless(task)) {
         invoke(forecast::Arima,
-          y = task$data(rows = task$row_ids, cols = task$target_names),
-          xreg = xreg,
+          y = as.ts(task$data(cols = task$target_names)),
           .args = pv
         )
       } else {
+        xreg = as.matrix(task$data(cols = task$feature_names))
         invoke(forecast::Arima,
-          y = task$data(rows = task$row_ids, cols = task$target_names),
-          .args = pv)
+          y = as.ts(task$data(cols = task$target_names)),
+          xreg = xreg,
+          .args = pv
+        )
       }
     },
 
     .predict = function(task) {
       pv = self$param_set$get_values(tags = "predict")
       if (private$.is_newdata(task)) {
-        if (length(task$feature_names) > 0) {
+        if (is_task_featureless(task)) {
+          prediction = invoke(forecast::forecast, self$model, h = length(task$row_ids))
+        } else {
           newdata = as.matrix(task$data(cols = task$feature_names))
           prediction = invoke(forecast::forecast, self$model, xreg = newdata)
-        } else {
-          prediction = invoke(forecast::forecast, self$model, h = length(task$row_ids))
-          browser()
         }
         list(response = prediction$mean)
       } else {
-        prediction = stats::fitted(self$model[task$row_ids])
+        prediction = stats::fitted(self$model)[task$row_ids]
         list(response = prediction)
       }
     },

diff --git a/R/LearnerRegrAutoARIMA.R b/R/LearnerRegrAutoARIMA.R
@@ -0,0 +1,105 @@
+#' @title ARIMA
+#'
+#' @name mlr_learners_fcst.auto_arima
+#'
+#' @description
+#' ...
+#'
+#' @templateVar id fcst.auto_arima
+#' @template learner
+#'
+#' @references
+#' ...
+#'
+#' @export
+#' @template seealso_learner
+LearnerFcstAutoARIMA = R6Class("LearnerFcstAutoARIMA",
+  inherit = LearnerRegr,
+  public = list(
+    #' @description
+    #' Creates a new instance of this [R6][R6::R6Class] class.
+    initialize = function() {
+      param_set = ps(
+        d          = p_int(0L, default = NA, tags = "train", special_vals = list(NA)),
+        D          = p_int(0L, default = NA, tags = "train", special_vals = list(NA)),
+        max.q      = p_int(0L, default = 5, tags = "train"),
+        max.p      = p_int(0L, default = 5, tags = "train"),
+        max.P      = p_int(0L, default = 2, tags = "train"),
+        max.Q      = p_int(0L, default = 2, tags = "train"),
+        max.order  = p_int(0L, default = 5, tags = "train"),
+        max.d      = p_int(0L, default = 2, tags = "train"),
+        max.D      = p_int(0L, default = 1, tags = "train"),
+        start.p    = p_int(0L, default = 2, tags = "train"),
+        start.q    = p_int(0L, default = 2, tags = "train"),
+        start.P    = p_int(0L, default = 2, tags = "train"),
+        start.Q    = p_int(0L, default = 2, tags = "train"),
+        stepwise   = p_lgl(default = FALSE, tags = "train"),
+        allowdrift = p_lgl(default = TRUE, tags = "train"),
+        seasonal   = p_lgl(default = FALSE, tags = "train")
+      )
+
+      super$initialize(
+        id = "fcst.auto_arima",
+        param_set = param_set,
+        feature_types = c("logical", "integer", "numeric"),
+        packages = c("mlr3forecast", "forecast"),
+        label = "Auto ARIMA",
+        man = "mlr3forecast::mlr_learners_fcst.arima"
+      )
+    }
+  ),
+
+  private = list(
+    .max_index = NULL,
+
+    .train = function(task) {
+      if ("ordered" %nin% task$properties) {
+        stopf("%s learner requires an ordered task.", self$id)
+      }
+      private$.max_index = max(task$data(cols = task$col_roles$order)[[1L]])
+      pv = self$param_set$get_values(tags = "train")
+      if ("weights" %in% task$properties) {
+        pv = insert_named(pv, list(weights = task$weights$weight))
+      }
+
+      if (is_task_featureless(task)) {
+        invoke(forecast::auto.arima,
+          y = as.ts(task$data(cols = task$target_names)),
+          .args = pv
+        )
+      } else {
+        xreg = as.matrix(task$data(cols = task$feature_names))
+        invoke(forecast::auto.arima,
+          y = as.ts(task$data(cols = task$target_names)),
+          xreg = xreg,
+          .args = pv
+        )
+      }
+    },
+
+    .predict = function(task) {
+      pv = self$param_set$get_values(tags = "predict")
+      if (private$.is_newdata(task)) {
+        if (is_task_featureless(task)) {
+          prediction = invoke(forecast::forecast, self$model, h = length(task$row_ids))
+        } else {
+          newdata = as.matrix(task$data(cols = task$feature_names))
+          prediction = invoke(forecast::forecast, self$model, xreg = newdata)
+        }
+        list(response = prediction$mean)
+      } else {
+        prediction = stats::fitted(self$model)[task$row_ids]
+        list(response = prediction)
+      }
+    },
+
+    .is_newdata = function(task) {
+      order_cols = task$col_roles$order
+      idx = task$backend$data(rows = task$row_ids, cols = order_cols)[[1L]]
+      !any(private$.max_index %in% idx)
+    }
+  )
+)
+
+#' @include zzz.R
+register_learner("fcst.auto_arima", LearnerFcstAutoARIMA)
diff --git a/R/utils.R b/R/utils.R
@@ -0,0 +1,4 @@
+is_task_featureless = function(task) {
+  length(task$feature_names) == 0L
+}
+
diff --git a/README.Rmd b/README.Rmd
@@ -285,13 +285,19 @@ newdata = generate_newdata(task, 12L, "month")
 newdata
 ```
 
-### Example: WIP
+### Example: Native Forecasting Learners
 
-```{r, eval = FALSE}
-task = tsk("airpassengers")
-task$select(setdiff(task$feature_names, "date"))
-learner = LearnerFcstARIMA$new()$train(task)
+```{r}
+task = tsk("airpassengers")$select(setdiff(task$feature_names, "date"))
+learner = lrn("fcst.arima", order = c(2L, 1L, 2L))$train(task)
+prediction = learner$predict(task, 140:144)
+prediction$score(msr("regr.rmse"))
+newdata = generate_newdata(task, 12L, "month")
+learner$predict_newdata(newdata, task)
+
+learner = lrn("fcst.auto_arima")$train(task)
+prediction = learner$predict(task, 140:144)
+prediction$score(msr("regr.rmse"))
 newdata = generate_newdata(task, 12L, "month")
-learner$predict(task, 140:144)
 learner$predict_newdata(newdata, task)
 ```