Merge pull request #301 from tidymodels/liquidSVM-changes

liquidSVM changes

Author: topepo

DESCRIPTION

@@ -50,4 +50,5 @@ Suggests:
     rpart,
     MASS,
     nlme,
-    modeldata
+    modeldata,
+    liquidSVM

NEWS.md

@@ -1,10 +1,18 @@
 # parsnip 0.1.1
 
+## New Features
+
+ * `liquidSVM` was added as an engine for `svm_rbf()` (#300)
+
+## Fixes
+
+* The error message for missing packages was fixed (#289 and #292)
+
+
 ## Other Changes
 
 * S3 dispatch for `tidy()` was broken on R 4.0. 
 
-* The error message for missing packages was fixed (#289 and #292)
 
 # parsnip 0.0.5
 

R/svm_rbf.R

@@ -36,11 +36,6 @@
 #' \item \pkg{R}:  `"liquidSVM"`
 #' }
 #'
-#' Note that models created using the `liquidSVM` engine cannot be saved like
-#' conventional R objects. The `fit` slot of the `model_fit` object has to be
-#' saved separately using the `liquidSVM::write.liquidSVM()` function. Likewise
-#' to restore a model, the `fit` slot has to be replaced with the model that is
-#' read using the `liquidSVM::read.liquidSVM()` function.
 #'
 #' @includeRmd man/rmd/svm-rbf.Rmd details
 #'

R/svm_rbf_data.R

@@ -243,9 +243,12 @@ set_pred(
   type = "prob",
   value = list(
     pre = function(x, object) {
-      if (object$fit$predict.prob == FALSE)
-        stop("`svm` model does not appear to use class probabilities. Was ",
-             "the model fit with `predict.prob = TRUE`?", call. = FALSE)
+      if (!object$fit$predict.prob) {
+        rlang::abort(
+          paste0("`svm` model does not appear to use class probabilities. Was ",
+                 "the model fit with `predict.prob = TRUE`?")
+        )
+      }
       x
     },
     post = function(result, object) {

man/rmd/svm-rbf.Rmd

@@ -19,6 +19,32 @@ svm_rbf() %>%
   translate()
 ```
 
+## liquidSVM
+
+```{r liquidSVM-reg}
+svm_rbf() %>% 
+  set_engine("liquidSVM") %>% 
+  set_mode("regression") %>% 
+  translate()
+```
+
+```{r liquidSVM-cls}
+svm_rbf() %>% 
+  set_engine("liquidSVM") %>% 
+  set_mode("classification") %>% 
+  translate()
+```
+
+Note that models created using the `liquidSVM` engine cannot be saved like
+conventional R objects. The `fit` slot of the `model_fit` object has to be saved
+separately using the `liquidSVM::write.liquidSVM()` function. Likewise to restore
+a model, the `fit` slot has to be replaced with the model that is read using the
+`liquidSVM::read.liquidSVM()` function.
+
+`liquidSVM` parameterizes the kernel parameter differently than `kernlab`. To 
+translate between engines, `sigma = 1/gammas^2`. Users will be specifying 
+`sigma` and the function translates the value to `gamma`. 
+
 ## Parameter translations
 
 The standardized parameter names in parsnip can be mapped to their original