update vignettes

R/pkg/R/mllib_regression.R

@@ -114,19 +114,18 @@ setMethod("spark.glm", signature(data = "SparkDataFrame", formula = "formula"),
               }
             }
             if (is.function(family)) {
-              # family = statmod::tweedie()
-              if (tolower(family$family) == "tweedie") {
-                family <- list(family = "tweedie", link = "linkNotUsed")
-                variancePower <- log(family$variance(exp(1)))
-                linkPower <- log(family$linkfun(exp(1)))
-              } else {
-                family <- family()
-              }
+              family <- family()
             }
             if (is.null(family$family)) {
               print(family)
               stop("'family' not recognized")
             }
+            # family = statmod::tweedie()
+            if (tolower(family$family) == "tweedie" && !is.null(family$variance)) {
+              variancePower <- log(family$variance(exp(1)))
+              linkPower <- log(family$linkfun(exp(1)))
+              family <- list(family = "tweedie", link = "linkNotUsed")
+            }
 
             formula <- paste(deparse(formula), collapse = "")
             if (!is.null(weightCol) && weightCol == "") {

R/pkg/inst/tests/testthat/test_mllib_regression.R

@@ -91,9 +91,9 @@ test_that("spark.glm and predict", {
   #' print(coef(rModel))
 
   rCoef <- c(0.6455409, 0.1169143, -0.3224752, -0.3282174)
-  rVals <- as.numeric(model.matrix(Sepal.Width ~ Sepal.Length + Species,
-                                   data = iris) %*% rCoef)
-  expect_true(all(abs(rVals - vals) < 1e-6), rVals - vals)
+  rVals <- exp(as.numeric(model.matrix(Sepal.Width ~ Sepal.Length + Species,
+                                       data = iris) %*% rCoef))
+  expect_true(all(abs(rVals - vals) < 1e-5), rVals - vals)
 
   # Test stats::predict is working
   x <- rnorm(15)
@@ -281,9 +281,9 @@ test_that("glm and predict", {
   #' print(coef(rModel))
 
   rCoef <- c(0.6455409, 0.1169143, -0.3224752, -0.3282174)
-  rVals <- as.numeric(model.matrix(Sepal.Width ~ Sepal.Length + Species,
-                                   data = iris) %*% rCoef)
-  expect_true(all(abs(rVals - vals) < 1e-6), rVals - vals)
+  rVals <- exp(as.numeric(model.matrix(Sepal.Width ~ Sepal.Length + Species,
+                                   data = iris) %*% rCoef))
+  expect_true(all(abs(rVals - vals) < 1e-5), rVals - vals)
 
   # Test stats::predict is working
   x <- rnorm(15)

R/pkg/vignettes/sparkr-vignettes.Rmd

@@ -682,7 +682,9 @@ There are three ways to specify the `family` argument.
 
 * Result returned by a family function, e.g. `family = poisson(link = log)`.
 
-* Note that when package `statmod` is loaded, the tweedie family is specified as `tweedie(var.power, link.power)`. Otherwise, one can use the SparkR internal definition `SparkR:::tweedie(var.power, link.power)`. In the above, `var.power` is the power index of the variance function and `link.power` is the index of the the power link function (the default value is `link.power = 1.0 - var.power`). This is consistent with the `tweedie` family defined in the `statmod` package.  Some examples: `family = tweedie(0.0)` is  gaussian with identity link, `family = tweedie(1.0)` poisson with log link, `family = tweedie(2.0)` Gamma with inverse link, and `family = tweedie(1.5, 0.0)` compound Poisson with log link.
+* Note that there are two ways to specify the tweedie family:
+  a) Set `family = "tweedie"` and specify the `variancePower` and `linkPower`
+  b) When package `statmod` is loaded, the tweedie family is specified using the family definition therein, i.e., `tweedie()`.
 
 For more information regarding the families and their link functions, see the Wikipedia page [Generalized Linear Model](https://en.wikipedia.org/wiki/Generalized_linear_model).
 
@@ -700,12 +702,13 @@ head(select(gaussianFitted, "model", "prediction", "mpg", "wt", "hp"))
 
 The following is the same fit using the tweedie family:
 ```{r}
-tweedieGLM1 <- spark.glm(carsDF, mpg ~ wt + hp, family = SparkR:::tweedie(0.0))
+tweedieGLM1 <- spark.glm(carsDF, mpg ~ wt + hp, family = "tweedie", variancePower = 0.0)
 summary(tweedieGLM1)
 ```
 We can try other distributions in the tweedie family, for example, a compound Poisson distribution with a log link:
 ```{r}
-tweedieGLM2 <- spark.glm(carsDF, mpg ~ wt + hp, family = SparkR:::tweedie(1.2, 0.0))
+tweedieGLM2 <- spark.glm(carsDF, mpg ~ wt + hp, family = "tweedie", 
+                         variancePower = 1.2, linkPower = 0.0)
 summary(tweedieGLM2)
 ```