Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[MINOR][SPARKR][ML] Joint coefficients with intercept for SparkR linear SVM summary. #18035

Closed
wants to merge 4 commits into from
Closed

[MINOR][SPARKR][ML] Joint coefficients with intercept for SparkR linear SVM summary. #18035

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
1ed3ba0
Code reorg and cleanup for SparkR linear SVM.
yanboliang May 19, 2017
39317c1
Update test case.
yanboliang May 19, 2017
3c14d15
Update docs.
yanboliang May 22, 2017
5d9afe0
Fix typo.
yanboliang May 23, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
38 changes: 15 additions & 23 deletions R/pkg/R/mllib_classification.R
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,15 +46,16 @@ setClass("MultilayerPerceptronClassificationModel", representation(jobj = "jobj"
#' @note NaiveBayesModel since 2.0.0
setClass("NaiveBayesModel", representation(jobj = "jobj"))

#' linear SVM Model
#' Linear SVM Model
#'
#' Fits an linear SVM model against a SparkDataFrame. It is a binary classifier, similar to svm in glmnet package
#' Fits a linear SVM model against a SparkDataFrame, similar to svm in e1071 package.
#' Currently only supports binary classification model with linear kernel.
#' Users can print, make predictions on the produced model and save the model to the input path.
#'
#' @param data SparkDataFrame for training.
#' @param formula A symbolic description of the model to be fitted. Currently only a few formula
#' operators are supported, including '~', '.', ':', '+', and '-'.
#' @param regParam The regularization parameter.
#' @param regParam The regularization parameter. Only supports L2 regularization currently.
#' @param maxIter Maximum iteration number.
#' @param tol Convergence tolerance of iterations.
#' @param standardization Whether to standardize the training features before fitting the model. The coefficients
Expand Down Expand Up @@ -111,10 +112,10 @@ setMethod("spark.svmLinear", signature(data = "SparkDataFrame", formula = "formu
new("LinearSVCModel", jobj = jobj)
})

# Predicted values based on an LinearSVCModel model
# Predicted values based on a LinearSVCModel model

#' @param newData a SparkDataFrame for testing.
#' @return \code{predict} returns the predicted values based on an LinearSVCModel.
#' @return \code{predict} returns the predicted values based on a LinearSVCModel.
#' @rdname spark.svmLinear
#' @aliases predict,LinearSVCModel,SparkDataFrame-method
#' @export
Expand All @@ -124,36 +125,27 @@ setMethod("predict", signature(object = "LinearSVCModel"),
predict_internal(object, newData)
})

# Get the summary of an LinearSVCModel
# Get the summary of a LinearSVCModel

#' @param object an LinearSVCModel fitted by \code{spark.svmLinear}.
#' @param object a LinearSVCModel fitted by \code{spark.svmLinear}.
#' @return \code{summary} returns summary information of the fitted model, which is a list.
#' The list includes \code{coefficients} (coefficients of the fitted model),
#' \code{intercept} (intercept of the fitted model), \code{numClasses} (number of classes),
#' \code{numFeatures} (number of features).
#' \code{numClasses} (number of classes), \code{numFeatures} (number of features).
#' @rdname spark.svmLinear
#' @aliases summary,LinearSVCModel-method
#' @export
#' @note summary(LinearSVCModel) since 2.2.0
setMethod("summary", signature(object = "LinearSVCModel"),
function(object) {
jobj <- object@jobj
features <- callJMethod(jobj, "features")
labels <- callJMethod(jobj, "labels")
coefficients <- callJMethod(jobj, "coefficients")
nCol <- length(coefficients) / length(features)
coefficients <- matrix(unlist(coefficients), ncol = nCol)
intercept <- callJMethod(jobj, "intercept")
features <- callJMethod(jobj, "rFeatures")
coefficients <- callJMethod(jobj, "rCoefficients")
coefficients <- as.matrix(unlist(coefficients))
colnames(coefficients) <- c("Estimate")
rownames(coefficients) <- unlist(features)
numClasses <- callJMethod(jobj, "numClasses")
numFeatures <- callJMethod(jobj, "numFeatures")
if (nCol == 1) {
Copy link
Contributor Author

@yanboliang yanboliang May 19, 2017 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ML LinearSVC only supports binary classification, and will not support multiple classification in the near future, so we can simplify here.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why not label, intercept? i think they are common in R to include what goes into the model (although in many cases it just include the formula in the model summary)

Copy link
Contributor Author

@yanboliang yanboliang May 22, 2017 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@felixcheung The change here is to make coefficients matrix has only one column named Estimate. I speculate the original code referred to spark.logit which supports multiple classification, so it should have multiple columns and each columns' name should be corresponding label. For binary classification, the coefficients are not bind to any labels, so we use Estimate as the column name like what R does. LinearSVC will not support multiple classification in the future, so I simplified it at here.
The followings are summary outputs for binomial and multinomial logistic regression in SparkR:
Binomial logistic regression model:
image
Multinomial logistic regression model:
image

colnames(coefficients) <- c("Estimate")
} else {
colnames(coefficients) <- unlist(labels)
}
rownames(coefficients) <- unlist(features)
list(coefficients = coefficients, intercept = intercept,
numClasses = numClasses, numFeatures = numFeatures)
list(coefficients = coefficients, numClasses = numClasses, numFeatures = numFeatures)
})

# Save fitted LinearSVCModel to the input path
Expand Down
3 changes: 1 addition & 2 deletions R/pkg/inst/tests/testthat/test_mllib_classification.R
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,9 +38,8 @@ test_that("spark.svmLinear", {
expect_true(class(summary$coefficients[, 1]) == "numeric")

coefs <- summary$coefficients[, "Estimate"]
expected_coefs <- c(-0.1563083, -0.460648, 0.2276626, 1.055085)
expected_coefs <- c(-0.06004978, -0.1563083, -0.460648, 0.2276626, 1.055085)
expect_true(all(abs(coefs - expected_coefs) < 0.1))
expect_equal(summary$intercept, -0.06004978, tolerance = 1e-2)

# Test prediction with string label
prediction <- predict(model, training)
Expand Down
12 changes: 10 additions & 2 deletions mllib/src/main/scala/org/apache/spark/ml/r/LinearSVCWrapper.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,9 +38,17 @@ private[r] class LinearSVCWrapper private (
private val svcModel: LinearSVCModel =
pipeline.stages(1).asInstanceOf[LinearSVCModel]

lazy val coefficients: Array[Double] = svcModel.coefficients.toArray
lazy val rFeatures: Array[String] = if (svcModel.getFitIntercept) {
Array("(Intercept)") ++ features
Copy link
Contributor Author

@yanboliang yanboliang May 19, 2017 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In R we stack intercept with other feature names, you can refer spark.glm, spark.logit, spark.survreg.

} else {
features
}

lazy val intercept: Double = svcModel.intercept
lazy val rCoefficients: Array[Double] = if (svcModel.getFitIntercept) {
Array(svcModel.intercept) ++ svcModel.coefficients.toArray
} else {
svcModel.coefficients.toArray
}

lazy val numClasses: Int = svcModel.numClasses

Expand Down