Over the past couple of years we’ve heard time and time again that people want a native dplyr interface to Spark, so we built one! sparklyr also provides interfaces to Spark’s distributed machine learning algorithms and much more. Highlights include:
Interactively manipulate Spark data using both dplyr and SQL (via DBI).
Filter and aggregate Spark datasets then bring them into R for analysis and visualization.
Create extensions that call the full Spark API and provide interfaces to Spark packages.
Integrated support for establishing Spark connections and browsing Spark data frames within the RStudio IDE.
We’re also excited to be working with several industry partners. IBM is incorporating sparklyr into their Data Science Experience, Cloudera is working with us to ensure that sparklyr meets the requirements of their enterprise customers, and H2O has provided an integration between sparklyr and H2O Sparkling Water.
You can install sparklyr from CRAN as follows:
You should also install a local version of Spark for development purposes:
library(sparklyr) spark_install(version = "1.6.2")
If you use the RStudio IDE, you should also download the latest preview release of the IDE which includes several enhancements for interacting with Spark.
Extensive documentation and examples are available at http://spark.rstudio.com.
Connecting to Spark
You can connect to both local instances of Spark as well as remote Spark clusters. Here we’ll connect to a local instance of Spark:
library(sparklyr) sc <- spark_connect(master = "local")
The returned Spark connection (
sc) provides a remote dplyr data source to the Spark cluster.
You can copy R data frames into Spark using the dplyr copy_to function (more typically though you’ll read data within the Spark cluster using the spark_read family of functions). For the examples below we’ll copy some datasets from R into Spark (note that you may need to install the nycflights13 and Lahman packages in order to execute this code):
library(dplyr) iris_tbl <- copy_to(sc, iris) flights_tbl <- copy_to(sc, nycflights13::flights, "flights") batting_tbl <- copy_to(sc, Lahman::Batting, "batting")
We can now use all of the available dplyr verbs against the tables within the cluster. Here’s a simple filtering example:
# filter by departure delay flights_tbl %>% filter(dep_delay == 2)
Introduction to dplyr provides additional dplyr examples you can try. For example, consider the last example from the tutorial which plots data on flight delays:
delay <- flights_tbl %>% group_by(tailnum) %>% summarise(count = n(), dist = mean(distance), delay = mean(arr_delay)) %>% filter(count > 20, dist < 2000, !is.na(delay)) %>% collect() # plot delays library(ggplot2) ggplot(delay, aes(dist, delay)) + geom_point(aes(size = count), alpha = 1/2) + geom_smooth() + scale_size_area(max_size = 2)
Note that while the dplyr functions shown above look identical to the ones you use with R data frames, with sparklyr they use Spark as their back end and execute remotely in the cluster.
dplyr window functions are also supported, for example:
batting_tbl %>% select(playerID, yearID, teamID, G, AB:H) %>% arrange(playerID, yearID, teamID) %>% group_by(playerID) %>% filter(min_rank(desc(H)) <= 2 & H > 0)
For additional documentation on using dplyr with Spark see the dplyr section of the sparklyr website.
It’s also possible to execute SQL queries directly against tables within a Spark cluster. The
spark_connection object implements a DBI interface for Spark, so you can use
dbGetQuery to execute SQL and return the result as an R data frame:
library(DBI) iris_preview <- dbGetQuery(sc, "SELECT * FROM iris LIMIT 10")
You can orchestrate machine learning algorithms in a Spark cluster via either Spark MLlib or via the H2O Sparkling Water extension package. Both provide a set of high-level APIs built on top of DataFrames that help you create and tune machine learning workflows.
In this example we’ll use ml_linear_regression to fit a linear regression model. We’ll use the built-in
mtcars dataset, and see if we can predict a car’s fuel consumption (
mpg) based on its weight (
wt) and the number of cylinders the engine contains (
cyl). We’ll assume in each case that the relationship between
mpg and each of our features is linear.
# copy mtcars into spark mtcars_tbl <- copy_to(sc, mtcars) # transform our data set, and then partition into 'training', 'test' partitions <- mtcars_tbl %>% filter(hp >= 100) %>% mutate(cyl8 = cyl == 8) %>% sdf_partition(training = 0.5, test = 0.5, seed = 1099) # fit a linear model to the training dataset fit <- partitions$training %>% ml_linear_regression(response = "mpg", features = c("wt", "cyl"))
For linear regression models produced by Spark, we can use
summary() to learn a bit more about the quality of our fit, and the statistical significance of each of our predictors.
Spark machine learning supports a wide array of algorithms and feature transformations, and as illustrated above it’s easy to chain these functions together with dplyr pipelines. To learn more see the Spark MLlib section of the sparklyr website.
H2O Sparkling Water
Let’s walk the same
mtcars example, but in this case use H2O’s machine learning algorithms via the H2O Sparkling Water extension. The dplyr code used to prepare the data is the same, but after partitioning into test and training data we call
h2o.glm rather than
# convert to h20_frame (uses the same underlying rdd) training <- as_h2o_frame(partitions$training) test <- as_h2o_frame(partitions$test) # fit a linear model to the training dataset fit <- h2o.glm(x = c("wt", "cyl"), y = "mpg", training_frame = training, lamda_search = TRUE) # inspect the model print(fit)
For linear regression models produced by H2O, we can use either
summary() to learn a bit more about the quality of our fit. The
summary() method returns some extra information about scoring history and variable importance.
To learn more see the H2O Sparkling Water section of the sparklyr website.
The facilities used internally by sparklyr for its dplyr and machine learning interfaces are available to extension packages. Since Spark is a general purpose cluster computing system there are many potential applications for extensions (e.g. interfaces to custom machine learning pipelines, interfaces to 3rd party Spark packages, etc.).
We’re excited to see what other sparklyr extensions the R community creates. To learn more see the Extensions section of the sparklyr website.
The latest RStudio Preview Release of the RStudio IDE includes integrated support for Spark and the sparklyr package, including tools for:
Creating and managing Spark connections
Browsing the tables and columns of Spark DataFrames
Previewing the first 1,000 rows of Spark DataFrames
Once you’ve installed the sparklyr package, you should find a new Spark pane within the IDE. This pane includes a New Connection dialog which can be used to make connections to local or remote Spark instances:
Once you’ve connected to Spark you’ll be able to browse the tables contained within the Spark cluster:
The Spark DataFrame preview uses the standard RStudio data viewer:
The RStudio IDE features for sparklyr are available now as part of the RStudio Preview Release. The final version of RStudio IDE that includes integrated support for sparklyr will ship within the next few weeks.
We’re very pleased to be joined in this announcement by IBM, Cloudera, and H2O, who are working with us to ensure that sparklyr meets the requirements of enterprise customers and is easy to integrate with current and future deployments of Spark.
“With our latest contributions to Apache Spark and the release of sparklyr, we continue to emphasize R as a primary data science language within the Spark community. Additionally, we are making plans to include sparklyr in Data Science Experience to provide the tools data scientists are comfortable with to help them bring business-changing insights to their companies faster,” said Ritika Gunnar, vice president of Offering Management, IBM Analytics.
“At Cloudera, data science is one of the most popular use cases we see for Apache Spark as a core part of the Apache Hadoop ecosystem, yet the lack of a compelling R experience has limited data scientists’ access to available data and compute,” said Charles Zedlewski, vice president, Products at Cloudera. “We are excited to partner with RStudio to help bring sparklyr to the enterprise, so that data scientists and IT teams alike can get more value from their existing skills and infrastructure, all with the security, governance, and management our customers expect.”
“At H2O.ai, we’ve been focused on bringing the best of breed open source machine learning to data scientists working in R & Python. However, the lack of robust tooling in the R ecosystem for interfacing with Apache Spark has made it difficult for the R community to take advantage of the distributed data processing capabilities of Apache Spark.
We’re excited to work with RStudio to bring the ease of use of dplyr and the distributed machine learning algorithms from H2O’s Sparkling Water to the R community via the sparklyr & rsparkling packages”