Merge pull request #316 from tidymodels/alt-argument-names

Handling alternative data argument names and fixing calls

Author: topepo

NAMESPACE

@@ -142,6 +142,7 @@ export(predict_quantile.model_fit)
 export(predict_raw)
 export(predict_raw.model_fit)
 export(rand_forest)
+export(repair_call)
 export(rpart_train)
 export(set_args)
 export(set_dependency)

NEWS.md

@@ -4,10 +4,20 @@
 
  * `tidyr` >= 1.0.0 is now required. 
 
+ * SVM models produced by `kernlab` now use the formula method. This change was due to how `ksvm()` made indicator variables for factor predictors (with one-hot encodings). Since the ordinary formula method did not do this, the data are passed as-is to `ksvm()` so that the results are closer to what one would get if `ksmv()` were called directly. 
+ 
+ * MARS models produced by `earth` now use the formula method. 
+ 
+ * Under-the-hood changes were made so that non-standard data arguments in the modeling packages can be accomodated. (#315)
+ 
 ## New Features
 
  * A new main argument was added to `boost_tree()` called `stop_iter` for early stopping. The `xgb_train()` function gained arguments for early stopping and a percentage of data to leave out for a validation set. 
 
+ * If `fit()` is used and the underlying model uses a formula, the _actual_ formula is pass to the model (instead of a placeholder). This makes the model call better. 
+ 
+ * A function named `repair_call()` was added. This can help change the underlying models `call` object to better reflect what they would have obtained if the model function had been used directly (instead of via `parsnip`). This is only useful when the user chooses a formula interface and the model uses a formula interface. It will also be of limited use when a recipes is used to construct the feature set in `workflows` or `tune`. 
+ 
 # parsnip 0.1.1
 
 ## New Features

R/aaa.R

@@ -66,7 +66,7 @@ utils::globalVariables(
     'lab', 'original', 'predicted_label', 'prediction', 'value', 'type',
     "neighbors", ".submodels", "has_submodel", "max_neighbor", "max_penalty",
     "max_terms", "max_tree", "model", "name", "num_terms", "penalty", "trees",
-    "sub_neighbors", ".pred_class")
+    "sub_neighbors", ".pred_class", "x", "y")
 )
 
 # nocov end

R/aaa_models.R

@@ -195,14 +195,35 @@ check_fit_info <- function(fit_obj) {
   if (is.null(fit_obj)) {
     rlang::abort("The `fit` module cannot be NULL.")
   }
+
+  # check required data elements
   exp_nms <- c("defaults", "func", "interface", "protect")
-  if (!isTRUE(all.equal(sort(names(fit_obj)), exp_nms))) {
+  has_req_nms <- exp_nms %in% names(fit_obj)
+
+  if (!all(has_req_nms)) {
     rlang::abort(
       glue::glue("The `fit` module should have elements: ",
       glue::glue_collapse(glue::glue("`{exp_nms}`"), sep = ", "))
       )
   }
 
+  # check optional data elements
+  opt_nms <- c("data")
+  other_nms <- setdiff(exp_nms, names(fit_obj))
+  has_opt_nms <- other_nms %in% opt_nms
+  if (any(!has_opt_nms)) {
+    msg <- glue::glue("The `fit` module can only have optional elements: ",
+                      glue::glue_collapse(glue::glue("`{exp_nms}`"), sep = ", "))
+
+    rlang::abort(msg)
+  }
+  if (any(other_nms == "data")) {
+    data_nms <- names(fit_obj$data)
+    if (length(data_nms == 0) || any(data_nms == "")) {
+      rlang::abort("All elements of the `data` argument vector must be named.")
+    }
+  }
+
   check_interface_val(fit_obj$interface)
   check_func_val(fit_obj$func)
 

R/arguments.R

@@ -24,7 +24,7 @@ check_eng_args <- function(args, obj, core_args) {
   if (length(common_args) > 0) {
     args <- args[!(names(args) %in% common_args)]
     common_args <- paste0(common_args, collapse = ", ")
-    rlang::warn(glue::glue("The following arguments cannot be manually modified",
+    rlang::warn(glue::glue("The following arguments cannot be manually modified ",
                            "and were removed: {common_args}."))
   }
   args
@@ -113,3 +113,94 @@ eval_args <- function(spec, ...) {
   spec$eng_args <- purrr::map(spec$eng_args, maybe_eval)
   spec
 }
+
+# ------------------------------------------------------------------------------
+
+# In some cases, a model function that we are calling has non-standard argument
+# names. For example, a function foo() that only has the x/y interface might
+# have a signature like `foo(X, Y)`.
+
+# To deal with this, we allow for the `data` element of the model
+# as an option to specify these actual argument names
+#
+#   value = list(
+#     interface = "xy",
+#     data = c(x = "X", y = "Y"),
+#     protect = c("X", "Y"),
+#     func = c(pkg = "bar", fun = "foo"),
+#     defaults = list()
+#   )
+
+make_call <- function(fun, ns, args, ...) {
+  # remove any null or placeholders (`missing_args`) that remain
+  discard <-
+    vapply(args, function(x)
+      is_missing_arg(x) | is.null(x), logical(1))
+  args <- args[!discard]
+
+  if (!is.null(ns) & !is.na(ns)) {
+    out <- call2(fun, !!!args, .ns = ns)
+  } else
+    out <- call2(fun, !!!args)
+  out
+}
+
+
+make_form_call <- function(object, env = NULL) {
+  fit_args <- object$method$fit$args
+
+  # Get the arguments related to data:
+  if (is.null(object$method$fit$data)) {
+    data_args <- c(formula = "formula", data = "data")
+  } else {
+    data_args <- object$method$fit$data
+  }
+
+  # add data arguments
+  for (i in seq_along(data_args)) {
+    fit_args[[ unname(data_args[i]) ]] <- sym(names(data_args)[i])
+  }
+
+  # sub in actual formula
+  fit_args[[ unname(data_args["formula"]) ]]  <- env$formula
+
+  if (object$engine == "spark") {
+    env$x <- env$data
+  }
+
+  fit_call <- make_call(
+    fun = object$method$fit$func["fun"],
+    ns = object$method$fit$func["pkg"],
+    fit_args
+  )
+  fit_call
+}
+
+make_xy_call <- function(object, target) {
+  fit_args <- object$method$fit$args
+
+  # Get the arguments related to data:
+  if (is.null(object$method$fit$data)) {
+    data_args <- c(x = "x", y = "y")
+  } else {
+    data_args <- object$method$fit$data
+  }
+
+  object$method$fit$args[[ unname(data_args["y"]) ]] <- rlang::expr(y)
+  object$method$fit$args[[ unname(data_args["x"]) ]] <-
+    switch(
+      target,
+      none = rlang::expr(x),
+      data.frame = rlang::expr(as.data.frame(x)),
+      matrix = rlang::expr(as.matrix(x)),
+      rlang::abort(glue::glue("Invalid data type target: {target}."))
+    )
+
+  fit_call <- make_call(
+    fun = object$method$fit$func["fun"],
+    ns = object$method$fit$func["pkg"],
+    object$method$fit$args
+  )
+
+  fit_call
+}

R/boost_tree_data.R

@@ -337,6 +337,7 @@ set_fit(
   mode = "regression",
   value = list(
     interface = "formula",
+    data = c(formula = "formula", data = "x"),
     protect = c("x", "formula", "type"),
     func = c(pkg = "sparklyr", fun = "ml_gradient_boosted_trees"),
     defaults = list(seed = expr(sample.int(10 ^ 5, 1)))
@@ -349,6 +350,7 @@ set_fit(
   mode = "classification",
   value = list(
     interface = "formula",
+    data = c(formula = "formula", data = "x"),
     protect = c("x", "formula", "type"),
     func = c(pkg = "sparklyr", fun = "ml_gradient_boosted_trees"),
     defaults = list(seed = expr(sample.int(10 ^ 5, 1)))

R/decision_tree_data.R

@@ -237,6 +237,7 @@ set_fit(
   mode = "regression",
   value = list(
     interface = "formula",
+    data = c(formula = "formula", data = "x"),
     protect = c("x", "formula"),
     func = c(pkg = "sparklyr", fun = "ml_decision_tree_classifier"),
     defaults =
@@ -250,6 +251,7 @@ set_fit(
   mode = "classification",
   value = list(
     interface = "formula",
+    data = c(formula = "formula", data = "x"),
     protect = c("x", "formula"),
     func = c(pkg = "sparklyr", fun = "ml_decision_tree_classifier"),
     defaults =

R/fit_helpers.R

@@ -30,21 +30,7 @@ form_form <-
     # sub in arguments to actual syntax for corresponding engine
     object <- translate(object, engine = object$engine)
 
-    fit_args <- object$method$fit$args
-
-    if (is_spark(object)) {
-      fit_args$x <- quote(x)
-      env$x <- env$data
-    } else {
-      fit_args$data <- quote(data)
-    }
-    fit_args$formula <- quote(formula)
-
-    fit_call <- make_call(
-      fun = object$method$fit$func["fun"],
-      ns = object$method$fit$func["pkg"],
-      fit_args
-    )
+    fit_call <- make_form_call(object, env = env)
 
     res <- list(
       lvl = y_levels,
@@ -89,21 +75,7 @@ xy_xy <- function(object, env, control, target = "none", ...) {
   # sub in arguments to actual syntax for corresponding engine
   object <- translate(object, engine = object$engine)
 
-  object$method$fit$args[["y"]] <- quote(y)
-  object$method$fit$args[["x"]] <-
-    switch(
-      target,
-      none = quote(x),
-      data.frame = quote(as.data.frame(x)),
-      matrix = quote(as.matrix(x)),
-      rlang::abort(glue::glue("Invalid data type target: {target}."))
-    )
-
-  fit_call <- make_call(
-    fun = object$method$fit$func["fun"],
-    ns = object$method$fit$func["pkg"],
-    object$method$fit$args
-  )
+  fit_call <- make_xy_call(object, target)
 
   res <- list(lvl = levels(env$y), spec = object)
 

R/linear_reg_data.R

@@ -312,6 +312,7 @@ set_fit(
   mode = "regression",
   value = list(
     interface = "formula",
+    data = c(formula = "formula", data = "x"),
     protect = c("x", "formula", "weight_col"),
     func = c(pkg = "sparklyr", fun = "ml_linear_regression"),
     defaults = list()

R/logistic_reg_data.R

@@ -236,6 +236,7 @@ set_fit(
   mode = "classification",
   value = list(
     interface = "formula",
+    data = c(formula = "formula", data = "x"),
     protect = c("x", "formula", "weight_col"),
     func = c(pkg = "sparklyr", fun = "ml_logistic_regression"),
     defaults =

R/mars_data.R

@@ -40,8 +40,8 @@ set_fit(
   eng = "earth",
   mode = "regression",
   value = list(
-    interface = "data.frame",
-    protect = c("x", "y", "weights"),
+    interface = "formula",
+    protect = c("formula", "data", "weights"),
     func = c(pkg = "earth", fun = "earth"),
     defaults = list(keepxy = TRUE)
   )
@@ -52,8 +52,8 @@ set_fit(
   eng = "earth",
   mode = "classification",
   value = list(
-    interface = "data.frame",
-    protect = c("x", "y", "weights"),
+    interface = "formula",
+    protect = c("formula", "data", "weights"),
     func = c(pkg = "earth", fun = "earth"),
     defaults = list(keepxy = TRUE)
   )

R/misc.R

@@ -115,22 +115,6 @@ convert_arg <- function(x) {
     x
 }
 
-make_call <- function(fun, ns, args, ...) {
-
-  #args <- map(args, convert_arg)
-
-  # remove any null or placeholders (`missing_args`) that remain
-  discard <-
-    vapply(args, function(x)
-      is_missing_arg(x) | is.null(x), logical(1))
-  args <- args[!discard]
-
-  if (!is.null(ns) & !is.na(ns)) {
-    out <- call2(fun, !!!args, .ns = ns)
-  } else
-    out <- call2(fun, !!!args)
-  out
-}
 
 levels_from_formula <- function(f, dat) {
   if (inherits(dat, "tbl_spark"))

R/multinom_reg_data.R

@@ -122,6 +122,7 @@ set_fit(
   mode = "classification",
   value = list(
     interface = "formula",
+    data = c(formula = "formula", data = "x"),
     protect = c("x", "formula", "weight_col"),
     func = c(pkg = "sparklyr", fun = "ml_logistic_regression"),
     defaults = list(family = "multinomial")

R/rand_forest_data.R

@@ -468,6 +468,7 @@ set_fit(
   mode = "classification",
   value = list(
     interface = "formula",
+    data = c(formula = "formula", data = "x"),
     protect = c("x", "formula", "type"),
     func = c(pkg = "sparklyr", fun = "ml_random_forest"),
     defaults = list(seed = expr(sample.int(10 ^ 5, 1)))
@@ -480,6 +481,7 @@ set_fit(
   mode = "regression",
   value = list(
     interface = "formula",
+    data = c(formula = "formula", data = "x"),
     protect = c("x", "formula", "type"),
     func = c(pkg = "sparklyr", fun = "ml_random_forest"),
     defaults = list(seed = expr(sample.int(10 ^ 5, 1)))

R/repair_call.R

@@ -0,0 +1,52 @@
+#' Repair a model call object
+#'
+#' When the user passes a formula to `fit()` _and_ the underyling model function
+#' uses a formula, the call object produced by `fit()` may not be usable by
+#' other functions. For example, some arguments may still be quosures and the
+#' `data` portion of the call will not correspond to the original data.
+#'
+#' `repair_call()` call can adjust the model objects call to be usable by other
+#' functions and methods.
+#' @param x A fitted `parsnip` model. An error will occur if the underlying model
+#'  does not have a `call` element.
+#' @param data A data object that is relavant to the call. In most cases, this
+#'  is the data frame that was given to `parsnip` for the model fit (i.e., the
+#'  training set data). The name of this data object is inserted into the call.
+#' @return A modified `parsnip` fitted model.
+#' @examples
+#'
+#' fitted_model <-
+#'   linear_reg() %>%
+#'   set_engine("lm", model = TRUE) %>%
+#'   fit(mpg ~ ., data = mtcars)
+#'
+#' # In this call, note that `data` is not `mtcars` and the `model = ~TRUE`
+#' # indicates that the `model` argument is an `rlang` quosure.
+#' fitted_model$fit$call
+#'
+#' # All better:
+#' repair_call(fitted_model, mtcars)$fit$call
+#' @export
+repair_call <- function(x, data) {
+  cl <- match.call()
+  if (!any(names(x$fit) == "call")) {
+    rlang::abort("No `call` object to modify.")
+  }
+  if (rlang::is_missing(data)) {
+    rlang::abort("Please supply a data object to `data`.")
+  }
+  fit_call <- x$fit$call
+  needs_eval <- purrr::map_lgl(fit_call, rlang::is_quosure)
+  if (any(needs_eval)) {
+    eval_args <- names(needs_eval)[needs_eval]
+    for(arg in eval_args) {
+      fit_call[[arg]] <- rlang::eval_tidy(fit_call[[arg]])
+    }
+  }
+  if (any(names(fit_call) == "data")) {
+    fit_call$data <- cl$data
+  }
+
+  x$fit$call <- fit_call
+  x
+}