Merge pull request #183 from tidymodels/knn-multi-predict

KNN multi predict

Author: topepo

NAMESPACE

@@ -10,6 +10,7 @@ S3method(multi_predict,"_earth")
 S3method(multi_predict,"_elnet")
 S3method(multi_predict,"_lognet")
 S3method(multi_predict,"_multnet")
+S3method(multi_predict,"_train.kknn")
 S3method(multi_predict,"_xgb.Booster")
 S3method(multi_predict,default)
 S3method(nullmodel,default)

NEWS.md

@@ -11,6 +11,7 @@
  * `add_rowindex()` can create a column called `.row` to a data frame. 
 
  * If a computational engine is not explicitly set, a default will be used. Each default is documented on the corresponding model page. A warning is issued at fit time unless verbosity is zero.  
+ * `nearest_neighbor` gained a `multi_predict` method. The `multi_predict()` documentation is a little better organized.  
 
 
 # parsnip 0.0.2

R/aaa.R

@@ -23,5 +23,6 @@ convert_stan_interval <- function(x, level = 0.95, lower = TRUE) {
 #' @importFrom utils globalVariables
 utils::globalVariables(
   c('.', '.label', '.pred', '.row', 'data', 'engine', 'engine2', 'group',
-    'lab', 'original', 'predicted_label', 'prediction', 'value', 'type')
+    'lab', 'original', 'predicted_label', 'prediction', 'value', 'type',
+    "neighbors")
   )

R/aaa_multi_predict.R

@@ -0,0 +1,35 @@
+# Define a generic to make multiple predictions for the same model object ------
+
+#' Model predictions across many sub-models
+#'
+#' For some models, predictions can be made on sub-models in the model object.
+#' @param object A `model_fit` object.
+#' @param new_data A rectangular data object, such as a data frame.
+#' @param type A single character value or `NULL`. Possible values
+#'  are "numeric", "class", "prob", "conf_int", "pred_int", "quantile",
+#'  or "raw". When `NULL`, `predict()` will choose an appropriate value
+#'  based on the model's mode.
+#' @param ... Optional arguments to pass to `predict.model_fit(type = "raw")`
+#'  such as `type`.
+#' @return A tibble with the same number of rows as the data being predicted.
+#'  Mostly likely, there is a list-column named `.pred` that is a tibble with
+#'  multiple rows per sub-model.
+#' @export
+multi_predict <- function(object, ...) {
+  if (inherits(object$fit, "try-error")) {
+    warning("Model fit failed; cannot make predictions.", call. = FALSE)
+    return(NULL)
+  }
+  UseMethod("multi_predict")
+}
+
+#' @export
+#' @rdname multi_predict
+multi_predict.default <- function(object, ...)
+  stop("No `multi_predict` method exists for objects with classes ",
+       paste0("'", class(), "'", collapse = ", "), call. = FALSE)
+
+#' @export
+predict.model_spec <- function(object, ...) {
+  stop("You must use `fit()` on your model specification before you can use `predict()`.", call. = FALSE)
+}

R/boost_tree.R

@@ -366,6 +366,8 @@ xgb_pred <- function(object, newdata, ...) {
 
 #' @importFrom purrr map_df
 #' @export
+#' @rdname multi_predict
+#' @param trees An integer vector for the number of trees in the ensemble.
 multi_predict._xgb.Booster <-
   function(object, new_data, type = NULL, trees = NULL, ...) {
     if (any(names(enquos(...)) == "newdata")) {
@@ -474,6 +476,7 @@ C5.0_train <-
   }
 
 #' @export
+#' @rdname multi_predict
 multi_predict._C5.0 <-
   function(object, new_data, type = NULL, trees = NULL, ...) {
     if (any(names(enquos(...)) == "newdata"))

R/linear_reg.R

@@ -302,6 +302,8 @@ predict_raw._elnet <- function(object, new_data, opts = list(), ...)  {
 #' @importFrom dplyr full_join as_tibble arrange
 #' @importFrom tidyr gather
 #' @export
+#'@rdname multi_predict
+#' @param penalty An numeric vector of penalty values.
 multi_predict._elnet <-
   function(object, new_data, type = NULL, penalty = NULL, ...) {
     if (any(names(enquos(...)) == "newdata"))

R/logistic_reg.R

@@ -276,6 +276,7 @@ predict._lognet <- function (object, new_data, type = NULL, opts = list(), penal
 #' @importFrom dplyr full_join as_tibble arrange
 #' @importFrom tidyr gather
 #' @export
+#' @rdname multi_predict
 multi_predict._lognet <-
   function(object, new_data, type = NULL, penalty = NULL, ...) {
     if (any(names(enquos(...)) == "newdata"))

R/mars.R

@@ -205,6 +205,8 @@ earth_reg_updater <- function(num, object, new_data, ...) {
 
 #' @importFrom purrr map_df
 #' @importFrom dplyr arrange
+#' @rdname multi_predict
+#' @param num_terms An integer vector for the number of MARS terms to retain.
 #' @export
 multi_predict._earth <-
   function(object, new_data, type = NULL, num_terms = NULL, ...) {

R/multinom_reg.R

@@ -232,6 +232,7 @@ predict._multnet <-
 #' @importFrom dplyr full_join as_tibble arrange
 #' @importFrom tidyr gather
 #' @export
+#' @rdname multi_predict
 multi_predict._multnet <-
   function(object, new_data, type = NULL, penalty = NULL, ...) {
     if (any(names(enquos(...)) == "newdata"))

R/nearest_neighbor.R

@@ -178,3 +178,43 @@ translate.nearest_neighbor <- function(x, engine = x$engine, ...) {
   }
   x
 }
+
+
+# ------------------------------------------------------------------------------
+
+#' @importFrom purrr map_df
+#' @importFrom dplyr starts_with
+#' @rdname multi_predict
+#' @param neighbors An integer vector for the number of nearest neighbors.
+#' @export
+multi_predict._train.kknn <-
+  function(object, new_data, type = NULL, neighbors = NULL, ...) {
+    if (any(names(enquos(...)) == "newdata"))
+      stop("Did you mean to use `new_data` instead of `newdata`?", call. = FALSE)
+
+    if (is.null(neighbors))
+      neighbors <- rlang::eval_tidy(object$fit$call$ks)
+    neighbors <- sort(neighbors)
+
+    if (is.null(type)) {
+      if (object$spec$mode == "classification")
+        type <- "class"
+      else
+        type <- "numeric"
+    }
+
+    res <-
+      purrr::map_df(neighbors, knn_by_k, object = object,
+                    new_data = new_data, type = type, ...)
+    res <- dplyr::arrange(res, .row, neighbors)
+    res <- split(res[, -1], res$.row)
+    names(res) <- NULL
+    dplyr::tibble(.pred = res)
+  }
+
+knn_by_k <- function(k, object, new_data, type, ...) {
+  object$fit$call$ks <- k
+  predict(object, new_data = new_data, type = type, ...) %>%
+    dplyr::mutate(neighbors = k, .row = dplyr::row_number()) %>%
+    dplyr::select(.row, neighbors, dplyr::starts_with(".pred"))
+}