SparkR is an R package that provides a light-weight frontend to use Apache Spark from R. SparkR also supports distributed machine learning using MLlib.
- For Spark 2.0 and above, you do not need to explicitly pass a
sqlContextobject to every function call. This article uses the new syntax. For old syntax examples, see SparkR 1.6 overview.
- For Spark 2.2 and above, notebooks no longer import SparkR by default because SparkR functions were conflicting with similarly named functions from other popular packages. To use SparkR you can call
library(SparkR)in your notebooks. The SparkR session is already configured, and all SparkR functions will talk to your attached cluster using the existing session.
You can run scripts that use SparkR on Databricks as spark-submit jobs, with minor code modifications. For an example, refer to Create and run a spark-submit job for R scripts.
You can create a DataFrame from a local R
data.frame, from a data source, or using a Spark SQL query.
The simplest way to create a DataFrame is to convert a local R
data.frame into a
SparkDataFrame. Specifically we can use
createDataFrame and pass in the local R
data.frame to create a
SparkDataFrame. Like most other SparkR functions,
syntax changed in Spark 2.0. You can see examples of this in the code snippet bellow.
Refer to createDataFrame for more examples.
library(SparkR) df <- createDataFrame(faithful) # Displays the content of the DataFrame to stdout head(df)
The general method for creating a DataFrame from a data source is
This method takes the path for the file to load and the type of data source.
SparkR supports reading CSV, JSON, text, and Parquet files
library(SparkR) diamondsDF <- read.df("/databricks-datasets/Rdatasets/data-001/csv/ggplot2/diamonds.csv", source = "csv", header="true", inferSchema = "true") head(diamondsDF)
SparkR automatically infers the schema from the CSV file.
Through Spark Packages you can find data source connectors for popular file formats such as Avro. As an example, use the spark-avro package to load an Avro file. The availability of the spark-avro package depends on your cluster’s image version. See Avro files.
First take an existing
data.frame, convert to a Spark DataFrame, and save it as an Avro file.
require(SparkR) irisDF <- createDataFrame(iris) write.df(irisDF, source = "com.databricks.spark.avro", path = "dbfs:/tmp/iris.avro", mode = "overwrite")
To verify that an Avro file was saved:
%fs ls /tmp/iris
Now use the spark-avro package again to read back the data.
irisDF2 <- read.df(path = "/tmp/iris.avro", source = "com.databricks.spark.avro") head(irisDF2)
The data source API can also be used to save DataFrames into
multiple file formats. For example, you can save the DataFrame from the
previous example to a Parquet file using
write.df(irisDF2, path="dbfs:/tmp/iris.parquet", source="parquet", mode="overwrite")
%fs ls dbfs:/tmp/people.parquet
You can also create SparkR DataFrames using Spark SQL queries.
# Register earlier df as temp view createOrReplaceTempView(people, "peopleTemp")
# Create a df consisting of only the 'age' column using a Spark SQL query age <- sql("SELECT age FROM peopleTemp")
age is a SparkDataFrame.
Spark DataFrames support a number of functions to do structured data processing. Here are some basic examples. A complete list can be found in the API docs.
# Import SparkR package if this is a new notebook require(SparkR) # Create DataFrame df <- createDataFrame(faithful)
# Select only the "eruptions" column head(select(df, df$eruptions))
# You can also pass in column name as strings head(select(df, "eruptions"))
# Filter the DataFrame to only retain rows with wait times shorter than 50 mins head(filter(df, df$waiting < 50))
SparkDataFrames support a number of commonly used functions to aggregate data after grouping. For example you can count the number of times each waiting time appears in the faithful dataset.
# You can also sort the output from the aggregation to get the most common waiting times waiting_counts <- count(groupBy(df, df$waiting)) head(arrange(waiting_counts, desc(waiting_counts$count)))
SparkR provides a number of functions that can be directly applied to columns for data processing and aggregation. The example below shows the use of basic arithmetic functions.
# Convert waiting time from hours to seconds. # You can assign this to a new column in the same DataFrame df$waiting_secs <- df$waiting * 60 head(df)
SparkR exposes most of MLLib algorithms. Under the hood, SparkR uses MLlib to train the model.
The example below shows the use of building a gaussian GLM model using
SparkR. To run linear regression, set family to
"gaussian". To run
logistic regression, set family to
"binomial". When using SparkML GLM SparkR
automatically performs one-hot encoding of
categorical features so that it does not need to be done manually.
Beyond String and Double type features, it is also possible to fit over
MLlib Vector features, for compatibility with other MLlib components.
# Create the DataFrame df <- createDataFrame(iris) # Fit a linear model over the dataset. model <- glm(Sepal_Length ~ Sepal_Width + Species, data = df, family = "gaussian") # Model coefficients are returned in a similar format to R's native glm(). summary(model)
For tutorials, see SparkR ML tutorials.