Methods inheriting from PipeOpImpute handle "POSIXct" and "Date"-features

DESCRIPTION

@@ -47,7 +47,11 @@ Authors@R:
       person(given = "Lona",
              family = "Koers",
              role = "ctb",
-             email = "[email protected]"))
+             email = "[email protected]"),
+      person(given = "Alexander",
+             family = "Winterstetter",
+             role = "ctb",
+             email = "[email protected]"))
 Description: Dataflow programming toolkit that enriches 'mlr3' with a diverse
   set of pipelining operators ('PipeOps') that can be composed into graphs.
   Operations exist for data preprocessing, model fitting, and ensemble

R/PipeOpImpute.R

@@ -329,7 +329,9 @@ PipeOpImpute = R6Class("PipeOpImpute",
         logical = c(TRUE, FALSE),
         numeric = 0, # see PipeOpImputeMean and PipeOpImputeMedian
         ordered = levels(feature),  # see above
-        character = ""
+        character = "",
+        Date = as.Date(0),
+        POSIXct = as.POSIXct(0)
       )
     },
 

R/PipeOpImputeConstant.R

@@ -83,7 +83,7 @@ PipeOpImputeConstant = R6Class("PipeOpImputeConstant",
         check_levels = p_lgl(init = TRUE, tags = c("train", "required"))
       )
       super$initialize(id, param_set = ps, param_vals = param_vals, empty_level_control = "always",
-        feature_types = c("logical", "integer", "numeric", "character", "factor", "ordered", "POSIXct"))
+        feature_types = c("logical", "integer", "numeric", "character", "factor", "ordered", "POSIXct", "Date"))
     }
   ),
   private = list(
@@ -96,7 +96,8 @@ PipeOpImputeConstant = R6Class("PipeOpImputeConstant",
         "character" = assert_string(constant),
         "factor"    = assert_string_or_factor(constant),
         "ordered"   = assert_string_or_factor(constant),
-        "POSIXct"   = assert_posixct(constant, any.missing = FALSE, len = 1L)
+        "POSIXct"   = assert_posixct(constant, any.missing = FALSE, len = 1L),
+        "Date" = assert_date(constant, any.missing = FALSE, len = 1L)
       )
       if (type %in% c("ordered", "factor") && self$param_set$values$check_levels) {
         if (!isTRUE(check_choice(as.character(constant), levels(feature)))) {

R/PipeOpImputeHist.R

@@ -1,11 +1,11 @@
-#' @title Impute Numerical Features by Histogram
+#' @title Impute Numeric, Integer, POSIXct or Date Features by Histogram
 #'
 #' @usage NULL
 #' @name mlr_pipeops_imputehist
 #' @format [`R6Class`][R6::R6Class] object inheriting from [`PipeOpImpute`]/[`PipeOp`].
 #'
 #' @description
-#' Impute numerical features by histogram.
+#' Impute numeric, integer, POSIXct or Date features by histogram.
 #'
 #' During training, a histogram is fitted on each column using R's [`hist()`][graphics::hist] function.
 #' The fitted histogram is then sampled from for imputation. Sampling happens in a two-step process:
@@ -27,7 +27,7 @@
 #' @section Input and Output Channels:
 #' Input and output channels are inherited from [`PipeOpImpute`].
 #'
-#' The output is the input [`Task`][mlr3::Task] with all affected numeric features missing values imputed by (column-wise) histogram; see Description for details.
+#' The output is the input [`Task`][mlr3::Task] with all affected numeric, integer, POSIXct or Date features missing values imputed by (column-wise) histogram; see Description for details.
 #'
 #' @section State:
 #' The `$state` is a named `list` with the `$state` elements inherited from [`PipeOpImpute`].
@@ -66,13 +66,27 @@ PipeOpImputeHist = R6Class("PipeOpImputeHist",
   inherit = PipeOpImpute,
   public = list(
     initialize = function(id = "imputehist", param_vals = list()) {
-      super$initialize(id, param_vals = param_vals, packages = "graphics", feature_types = c("integer", "numeric"))
+      super$initialize(id, param_vals = param_vals, packages = "graphics", feature_types = c("integer", "numeric", "Date", "POSIXct"))
     }
   ),
   private = list(
 
     .train_imputer = function(feature, type, context) {
-      graphics::hist(feature, plot = FALSE)[c("counts", "breaks")]
+      if (inherits(feature, c("POSIXct", "Date"))) {
+        # hist() for POSIXct/Date does not do "Sturges" breaks automatically, so we compute it explicitly
+        n_breaks = ceiling(log2(length(feature)) + 1)
+        # If we pass the number of breaks, hist() does some computation that results in integer overflow
+        if (inherits(feature, "POSIXct")) {
+          breaks = as.POSIXct(as.numeric(pretty(range(feature, na.rm = TRUE), n = n_breaks, min.n = 1)))
+        } else {
+          breaks = as.Date(as.numeric(pretty(range(feature, na.rm = TRUE), n = n_breaks, min.n = 1)))
+        }
+        # pretty() does not return values of length < 2, so the special case where `breaks` gets
+        # intepreted differently does not need to be handled here.
+        graphics::hist(feature, breaks = breaks, plot = FALSE)[c("counts", "breaks")]
+      } else {
+        graphics::hist(feature, plot = FALSE)[c("counts", "breaks")]
+      }
     },
 
     .impute = function(feature, type, model, context) {

R/PipeOpImputeLearner.R

@@ -10,7 +10,7 @@
 #' Note this parameter is part of the [`PipeOpImpute`] base class and explained there.
 #'
 #' Additionally, only features supported by the learner can be imputed; i.e. learners of type
-#' `regr` can only impute features of type `integer` and `numeric`, while `classif` can impute
+#' `regr` can only impute features of type `integer`, `numeric`, `POSIXct` and `Date`, while `classif` can impute
 #' features of type `factor`, `ordered` and `logical`.
 #'
 #' The [`Learner`][mlr3::Learner] used for imputation is trained on all `context_columns`; if these contain missing values,
@@ -105,7 +105,7 @@ PipeOpImputeLearner = R6Class("PipeOpImputeLearner",
       private$.learner = as_learner(learner, clone = TRUE)
       id = id %??% private$.learner$id
       feature_types = switch(private$.learner$task_type,
-        regr = c("integer", "numeric"),
+        regr = c("integer", "numeric", "POSIXct", "Date"),
         classif = c("logical", "factor", "ordered"),
         stop("Only `classif` or `regr` Learners are currently supported by PipeOpImputeLearner.")
         # FIXME: at least ordinal should also be possible. When Moore's law catches up with us we could even do `character`
@@ -183,6 +183,8 @@ PipeOpImputeLearner = R6Class("PipeOpImputeLearner",
       # Convert non-factor imputation targets to a factor
       if (is.numeric(feature)) {
         feature
+      } else if (any(class(feature) %in% c("POSIXct", "Date"))) {
+        as.numeric(feature)
       } else {
         if (!is.null(levels(feature))) {
           factor(feature, levels = levels(feature), ordered = FALSE)
@@ -198,6 +200,8 @@ PipeOpImputeLearner = R6Class("PipeOpImputeLearner",
         feature = round(feature)
       }
       if (type == "logical") feature = as.logical(feature) # FIXME mlr-org/mlr3#475
+      if (type == "POSIXct") feature = as.POSIXct(feature)
+      if (type == "Date") feature = as.Date(feature)
       auto_convert(feature, "feature to be imputed", type, levels = levels(feature))
     },
     .additional_phash_input = function() private$.learner$phash

R/PipeOpImputeMean.R

@@ -1,11 +1,11 @@
-#' @title Impute Numerical Features by their Mean
+#' @title Impute Numeric, Integer, POSIXct or Date Features by their Mean
 #'
 #' @usage NULL
 #' @name mlr_pipeops_imputemean
 #' @format [`R6Class`][R6::R6Class] object inheriting from [`PipeOpImpute`]/[`PipeOp`].
 #'
 #' @description
-#' Impute numerical features by their mean.
+#' Impute numeric, integer, POSIXct or Date features by their mean.
 #'
 #' @section Construction:
 #' ```
@@ -20,12 +20,12 @@
 #' @section Input and Output Channels:
 #' Input and output channels are inherited from [`PipeOpImpute`].
 #'
-#' The output is the input [`Task`][mlr3::Task] with all affected numeric features missing values imputed by (column-wise) mean.
+#' The output is the input [`Task`][mlr3::Task] with all affected numeric, integer, POSIXct and Date features missing values imputed by (column-wise) mean.
 #'
 #' @section State:
 #' The `$state` is a named `list` with the `$state` elements inherited from [`PipeOpImpute`].
 #'
-#' The `$state$model` is a named `list` of `numeric(1)` indicating the mean of the respective feature.
+#' The `$state$model` is a named `list` of either `numeric(1)`, `integer(1)`, `POSIXct(1)` or  `Date(1)` indicating the mean of the respective feature.
 #'
 #' @section Parameters:
 #' The parameters are the parameters inherited from [`PipeOpImpute`].
@@ -59,7 +59,7 @@ PipeOpImputeMean = R6Class("PipeOpImputeMean",
   inherit = PipeOpImpute,
   public = list(
     initialize = function(id = "imputemean", param_vals = list()) {
-      super$initialize(id, param_vals = param_vals, feature_types= c("numeric", "integer"))
+      super$initialize(id, param_vals = param_vals, feature_types= c("numeric", "integer", "POSIXct", "Date"))
     }
   ),
   private = list(

R/PipeOpImputeMedian.R

@@ -1,11 +1,11 @@
-#' @title Impute Numerical Features by their Median
+#' @title Impute Numeric, Integer, POSIXct or Date Features by their Median
 #'
 #' @usage NULL
 #' @name mlr_pipeops_imputemedian
 #' @format [`R6Class`][R6::R6Class] object inheriting from [`PipeOpImpute`]/[`PipeOp`].
 #'
 #' @description
-#' Impute numerical features by their median.
+#' Impute numerical, integer, POSIXct or Date features by their median.
 #'
 #' @section Construction:
 #' ```
@@ -20,12 +20,12 @@
 #' @section Input and Output Channels:
 #' Input and output channels are inherited from [`PipeOpImpute`].
 #'
-#' The output is the input [`Task`][mlr3::Task] with all affected numeric features missing values imputed by (column-wise) median.
+#' The output is the input [`Task`][mlr3::Task] with all affected numeric, integer, POSIXct and Date features missing values imputed by (column-wise) median.
 #'
 #' @section State:
 #' The `$state` is a named `list` with the `$state` elements inherited from [`PipeOpImpute`].
 #'
-#' The `$state$model` is a named `list` of `numeric(1)` indicating the median of the respective feature.
+#' The `$state$model` is a named `list` of `numeric(1)`, `integer(1)`, `POSIXct(1)` or  `Date(1)` indicating the median of the respective feature.
 #'
 #' @section Parameters:
 #' The parameters are the parameters inherited from [`PipeOpImpute`].
@@ -59,7 +59,7 @@ PipeOpImputeMedian = R6Class("PipeOpImputeMedian",
   inherit = PipeOpImpute,
   public = list(
     initialize = function(id = "imputemedian", param_vals = list()) {
-      super$initialize(id, param_vals = param_vals, packages = "stats", feature_types = c("numeric", "integer"))
+      super$initialize(id, param_vals = param_vals, packages = "stats", feature_types = c("numeric", "integer", "POSIXct", "Date"))
     }
   ),
   private = list(

R/PipeOpImputeMode.R

@@ -5,7 +5,7 @@
 #' @format [`R6Class`][R6::R6Class] object inheriting from [`PipeOpImpute`]/[`PipeOp`].
 #'
 #' @description
-#' Impute features by their mode. Supports factors as well as logical and numerical features.
+#' Impute features by their mode. Supports factors, logical, numerical, POSIXct and Date features.
 #' If multiple modes are present then imputed values are sampled randomly from them.
 #'
 #' @section Construction:
@@ -66,7 +66,7 @@ PipeOpImputeMode = R6Class("PipeOpImputeMode",
   inherit = PipeOpImpute,
   public = list(
     initialize = function(id = "imputemode", param_vals = list()) {
-      super$initialize(id, param_vals = param_vals, feature_types = c("factor", "integer", "logical", "numeric", "ordered"))
+      super$initialize(id, param_vals = param_vals, feature_types = c("factor", "integer", "logical", "numeric", "ordered", "POSIXct", "Date"))
     }
   ),
   private = list(

R/PipeOpImputeOOR.R

@@ -7,7 +7,7 @@
 #' @description
 #' Impute factorial features by adding a new level `".MISSING"`.
 #'
-#' Impute numerical features by constant values shifted below the minimum or above the maximum by
+#' Impute numeric, integer, POSIXct or Date features by constant values shifted below the minimum or above the maximum by
 #' using \eqn{min(x) - offset - multiplier * diff(range(x))} or
 #' \eqn{max(x) + offset + multiplier * diff(range(x))}.
 #'
@@ -51,25 +51,25 @@
 #'
 #' The `$state$model` contains either `".MISSING"` used for `character` and `factor` (also
 #' `ordered`) features or `numeric(1)` indicating the constant value used for imputation of
-#' `integer` and `numeric` features.
+#' `integer`, `numeric`, `POSIXct` or `Date` features.
 #'
 #' @section Parameters:
 #' The parameters are the parameters inherited from [`PipeOpImpute`], as well as:
 #' * `min` :: `logical(1)` \cr
 #'   Should `integer` and `numeric` features be shifted below the minimum? Initialized to `TRUE`. If `FALSE`
 #'   they are shifted above the maximum. See also the description above.
 #' * `offset` :: `numeric(1)` \cr
-#'   Numerical non-negative offset as used in the description above for `integer` and `numeric`
+#'   Numerical non-negative offset as used in the description above for `integer`, `numeric`, `POSIXCT` and `Date`.
 #'   features. Initialized to `1`.
 #' * `multiplier` :: `numeric(1)` \cr
-#'   Numerical non-negative multiplier as used in the description above for `integer` and `numeric`
+#'   Numerical non-negative multiplier as used in the description above for `integer`, `numeric`, `POSIXct` and `Date`.
 #'   features. Initialized to `1`.
 #'
 #' @section Internals:
 #' Adds an explicit new `level()` to `factor` and `ordered` features, but not to `character` features.
 #' For `integer` and `numeric` features uses the `min`, `max`, `diff` and `range` functions.
 #' `integer` and `numeric` features that are entirely `NA` are imputed as `0`. `factor` and `ordered` features that are
-#' entirely `NA` are imputed as `".MISSING"`.
+#' entirely `NA` are imputed as `".MISSING"`. For `POSIXct` and `Date` features the value `0` is transformed into the respective data type.
 #'
 #' @section Fields:
 #' Only fields inherited from [`PipeOp`].
@@ -119,7 +119,7 @@ PipeOpImputeOOR = R6Class("PipeOpImputeOOR",
       )
       # this is one of the few imputers that handles 'character' features!
       super$initialize(id, param_set = ps, param_vals = param_vals, empty_level_control = "param",
-        feature_types = c("character", "factor", "integer", "numeric", "ordered"))
+        feature_types = c("character", "factor", "integer", "numeric", "ordered", "POSIXct", "Date"))
     }
   ),
   private = list(
@@ -153,10 +153,13 @@ PipeOpImputeOOR = R6Class("PipeOpImputeOOR",
         logical = c(TRUE, FALSE),
         numeric = 0,
         ordered = ".MISSING",
-        character = ""
+        character = "",
+        POSIXct = as.POSIXct(0),
+        Date = as.Date(0)
       )
     }
   )
 )
 
 mlr_pipeops$add("imputeoor", PipeOpImputeOOR)
+

R/PipeOpImputeSample.R

@@ -33,8 +33,8 @@
 #' @section Internals:
 #' Uses the `sample()` function. Features that are entirely `NA` are imputed as
 #' the following: For `factor` or `ordered`, random levels are sampled uniformly at random.
-#' For logicals, `TRUE` or `FALSE` are sampled uniformly at random.
-#' Numerics and integers are imputed as `0`.
+#' For `logical`, `TRUE` or `FALSE` are sampled uniformly at random.
+#' `numeric` and `integer` are imputed as `0`.
 #'
 #' @section Fields:
 #' Only fields inherited from [`PipeOp`].
@@ -61,7 +61,7 @@ PipeOpImputeSample = R6Class("PipeOpImputeSample",
   inherit = PipeOpImpute,
   public = list(
     initialize = function(id = "imputesample", param_vals = list()) {
-      super$initialize(id, param_vals = param_vals, feature_types = c("factor", "integer", "logical", "numeric", "ordered"))
+      super$initialize(id, param_vals = param_vals, feature_types = c("factor", "integer", "logical", "numeric", "ordered", "POSIXct", "Date"))
     }
   ),
   private = list(

R/PipeOpTaskPreproc.R

@@ -59,7 +59,7 @@
 #'   `"TaskRegr"` (or another subclass introduced by other packages). Default is `"Task"`.
 #' * `tags` :: `character` | `NULL`\cr
 #'   Tags of the resulting `PipeOp`. This is added to the tag `"data transform"`. Default `NULL`.
-#'* `feature_types` :: `character`\cr
+#' * `feature_types` :: `character`\cr
 #'   Feature types affected by the `PipeOp`. See `private$.select_cols()` for more information.
 #'   Defaults to all available feature types.
 #'