README.Rmd

---
title: "sparklyr: R interface for Apache Spark"
output:
  github_document:
    fig_width: 9
    fig_height: 5
---

<!-- README.md is generated from README.Rmd. Please edit that file -->

```{r, include = FALSE}
knitr::opts_chunk$set(
  eval = TRUE, 
  warning = FALSE,
  collapse = TRUE,
  comment = "#>",
  fig.path = "tools/readme/", 
  dev = "png",
  out.width = "100%"
)

options(width = 60)

library(sparklyr)
library(dplyr)
library(ggplot2)
```

<!-- badges: start -->
[![R-CMD-check](https://github.com/sparklyr/sparklyr/actions/workflows/R-CMD-check.yaml/badge.svg)](https://github.com/sparklyr/sparklyr/actions/workflows/R-CMD-check.yaml)
[![Spark-Tests](https://github.com/sparklyr/sparklyr/actions/workflows/spark-tests.yaml/badge.svg)](https://github.com/sparklyr/sparklyr/actions/workflows/spark-tests.yaml)
[![CRAN status](https://www.r-pkg.org/badges/version/sparklyr)](https://CRAN.R-project.org/package=sparklyr)
[![Codecov test coverage](https://codecov.io/gh/sparklyr/sparklyr/branch/main/graph/badge.svg)](https://app.codecov.io/gh/sparklyr/sparklyr?branch=main)
<!-- badges: end -->

<img src="tools/readme/sparklyr-diagram.png" width="320" align="right" style="margin-left: 20px; margin-right: 20px"/>


- Install and connect to [Spark](https://spark.apache.org/) using YARN, Mesos,
Livy or Kubernetes.
- Use [dplyr](#using-dplyr) to filter and aggregate Spark datasets and 
[streams](https://spark.posit.co/guides/streaming/) then bring them into R 
for analysis and visualization.
- Use [MLlib](#machine-learning), [H2O](#using-h2o), 
[XGBoost](https://spark.posit.co/packages/sparkxgb/latest/) and
[GraphFrames](https://spark.posit.co/packages/graphframes/latest/) to train
models at scale in Spark.
- Create interoperable machine learning 
[pipelines](https://spark.posit.co/guides/pipelines.html) and productionize
them with [MLeap](https://spark.posit.co/packages/mleap/latest/).
- Create [extensions](#extensions) that call the full Spark API or run
[distributed R](#distributed-r) code to support new functionality.

## Table of Contents

```{r, eval = FALSE, echo = FALSE}
toc <- function() {
  re <- readLines("README.Rmd")
  has_title <- as.logical(lapply(re, function(x) substr(x, 1, 2) == "##"))
  only_titles <- re[has_title]
  titles <- trimws(gsub("#", "", only_titles))
  links <- trimws(gsub("`", "", titles))
  links <- tolower(links)
  links <- trimws(gsub(" ", "-", links))
  links <- trimws(gsub(",", "", links))
  toc_list <- lapply(
    seq_along(titles),
    function(x) {
      pad <- ifelse(substr(only_titles[x], 1, 3) == "###", "    - ", "  - ")
      paste0(pad, "[", titles[x], "](#",links[x], ")")
    }
  )
  toc_full <- paste(toc_list[2:length(toc_list)], collapse = "\n") 
  cat(toc_full)
}
toc()
```
  - [Installation](#installation)
  - [Connecting to Spark](#connecting-to-spark)
  - [Using dplyr](#using-dplyr)
    - [Window Functions](#window-functions)
  - [Using SQL](#using-sql)
  - [Machine Learning](#machine-learning)
  - [Reading and Writing Data](#reading-and-writing-data)
  - [Distributed R](#distributed-r)
  - [Extensions](#extensions)
  - [Table Utilities](#table-utilities)
  - [Connection Utilities](#connection-utilities)
  - [RStudio IDE](#rstudio-ide)
  - [Using H2O](#using-h2o)
  - [Connecting through Livy](#connecting-through-livy)
  - [Connecting through Databricks Connect](#connecting-through-databricks-connect-v2)
  
## Installation

You can install the **sparklyr** package from [CRAN](https://CRAN.r-project.org) as follows:

```{r, eval=FALSE}
install.packages("sparklyr")
```

You should also install a local version of Spark for development purposes:

```{r, eval=FALSE}
library(sparklyr)
spark_install()
```

To upgrade to the latest version of sparklyr, run the following command and restart your r session:

```{r, eval=FALSE}
install.packages("devtools")
devtools::install_github("sparklyr/sparklyr")
```

## 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 via the 
[spark_connect](https://spark.posit.co/packages/sparklyr/latest/reference/spark-connections.html) function:

```{r sparklyr-connect, message=FALSE}
library(sparklyr)
sc <- spark_connect(master = "local")
```

The returned Spark connection (`sc`) provides a remote dplyr data source to the Spark cluster.

For more information on connecting to remote Spark clusters see the 
[Deployment](https://spark.posit.co/deployment.html) section of the sparklyr
website.

## Using dplyr

We can now use all of the available dplyr verbs against the tables within the cluster.

We'll start by copying some datasets from R into the Spark cluster (note that you
may need to install the nycflights13 and Lahman packages in order to execute this code):

```{r, eval=FALSE}
install.packages(c("nycflights13", "Lahman"))
```

```{r dplyr-copy, message=FALSE}
library(dplyr)
iris_tbl <- copy_to(sc, iris, overwrite = TRUE)
flights_tbl <- copy_to(sc, nycflights13::flights, "flights", overwrite = TRUE)
batting_tbl <- copy_to(sc, Lahman::Batting, "batting", overwrite = TRUE)
src_tbls(sc)
```

To start with here's a simple filtering example:

```{r dplyr-filter}
# filter by departure delay and print the first few records
flights_tbl %>% filter(dep_delay == 2)
```

[Introduction to dplyr](https://spark.posit.co/guides/dplyr.html) provides
additional `dplyr` examples you can try. For example, consider the last example
from the tutorial which plots data on flight delays:

```{r dplyr-ggplot2}
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)
```


### Window Functions

dplyr [window functions](https://spark.posit.co/guides/dplyr.html#grouping)
are also supported, for example:

```{r dplyr-window}
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](https://spark.posit.co/dplyr.html) section of the sparklyr website.

## Using SQL

It's also possible to execute SQL queries directly against tables within a Spark
cluster. The `spark_connection` object implements a [DBI](https://github.com/r-dbi/DBI)
interface for Spark, so you can use `dbGetQuery()` to execute SQL and return the
result as an R data frame:

```{r sql-dbi}
library(DBI)
iris_preview <- dbGetQuery(sc, "SELECT * FROM iris LIMIT 10")
iris_preview
```

## Machine Learning

You can orchestrate machine learning algorithms in a Spark cluster via the
[machine learning](https://spark.apache.org/docs/latest/mllib-guide.html)
functions within **sparklyr**. These functions connect to a set of high-level
APIs built on top of DataFrames that help you create and tune machine learning
workflows.

Here's an example where we use [ml_linear_regression](https://spark.posit.co/packages/sparklyr/latest/reference/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.

```{r}
# copy mtcars into spark
mtcars_tbl <- copy_to(sc, mtcars, overwrite = TRUE)

# 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"))
fit
```

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.

```{r}
summary(fit)
```

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
[machine learning](https://spark.posit.co/mlib/) section.

## Reading and Writing Data

You can read and write data in CSV, JSON, and Parquet formats. Data can be stored in HDFS, S3, or on the local filesystem of cluster nodes.

```{r}
temp_csv <- tempfile(fileext = ".csv")
temp_parquet <- tempfile(fileext = ".parquet")
temp_json <- tempfile(fileext = ".json")

spark_write_csv(iris_tbl, temp_csv)
iris_csv_tbl <- spark_read_csv(sc, "iris_csv", temp_csv)

spark_write_parquet(iris_tbl, temp_parquet)
iris_parquet_tbl <- spark_read_parquet(sc, "iris_parquet", temp_parquet)

spark_write_json(iris_tbl, temp_json)
iris_json_tbl <- spark_read_json(sc, "iris_json", temp_json)

src_tbls(sc)
```


## Distributed R

You can execute arbitrary r code across your cluster using `spark_apply()`. For
example, we can apply `rgamma` over `iris` as follows:

```{r}
spark_apply(iris_tbl, function(data) {
  data[1:4] + rgamma(1,2)
})
```

You can also group by columns to perform an operation over each group of rows
and make use of any package within the closure:

```{r}
spark_apply(
  iris_tbl,
  function(e) broom::tidy(lm(Petal_Width ~ Petal_Length, e)),
  columns = c("term", "estimate", "std.error", "statistic", "p.value"),
  group_by = "Species"
)
```

## Extensions

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.).

Here's a simple example that wraps a Spark text file line counting function with 
an R function:

```{r}
# write a CSV
tempfile <- tempfile(fileext = ".csv")
write.csv(nycflights13::flights, tempfile, row.names = FALSE, na = "")

# define an R interface to Spark line counting
count_lines <- function(sc, path) {
  spark_context(sc) %>%
    invoke("textFile", path, 1L) %>%
      invoke("count")
}

# call spark to count the lines of the CSV
count_lines(sc, tempfile)
```


To learn more about creating extensions see the 
[Extensions](https://spark.posit.co/guides/extensions.html) section of the 
sparklyr website.


## Table Utilities

You can cache a table into memory with:

```{r, eval=FALSE}
tbl_cache(sc, "batting")
```

and unload from memory using:

```{r, eval=FALSE}
tbl_uncache(sc, "batting")
```


## Connection Utilities

You can view the Spark web console using the `spark_web()` function:

```{r, eval=FALSE}
spark_web(sc)
```

You can show the log using the `spark_log()` function:

```{r}
spark_log(sc, n = 10)
```

Finally, we disconnect from Spark:

```{r}
  spark_disconnect(sc)
```

## RStudio IDE

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:

<img src="tools/readme/spark-connect.png" class="screenshot" width=389 />

Once you've connected to Spark you'll be able to browse the tables contained 
within the Spark cluster and preview Spark DataFrames using the standard RStudio 
data viewer:

<img src="tools/readme/spark-dataview.png" class="screenshot" width=639 />

You can also connect to Spark through [Livy](https://livy.apache.org/) through 
a new connection dialog:

<img src="tools/readme/spark-connect-livy.png" class="screenshot" width=389 />

<div style="margin-bottom: 15px;"></div>


## Using H2O

[rsparkling](https://cran.r-project.org/package=rsparkling) is a CRAN package 
from [H2O](https://h2o.ai/) that extends [sparklyr](https://spark.posit.co/) 
to provide an interface into [Sparkling Water](https://github.com/h2oai/sparkling-water). 
For instance, the following example installs, configures and runs
[h2o.glm](https://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/glm.html):

```{r results='hide', message=FALSE, eval=FALSE}
library(rsparkling)
library(sparklyr)
library(dplyr)
library(h2o)

sc <- spark_connect(master = "local", version = "2.3.2")
mtcars_tbl <- copy_to(sc, mtcars, "mtcars", overwrite = TRUE)

mtcars_h2o <- as_h2o_frame(sc, mtcars_tbl, strict_version_check = FALSE)

mtcars_glm <- h2o.glm(x = c("wt", "cyl"),
                      y = "mpg",
                      training_frame = mtcars_h2o,
                      lambda_search = TRUE)
```

```{r eval=FALSE}
mtcars_glm
```
```
#> Model Details:
#> ==============
#>
#> H2ORegressionModel: glm
#> Model ID:  GLM_model_R_1527265202599_1
#> GLM Model: summary
#>     family     link                              regularization
#> 1 gaussian identity Elastic Net (alpha = 0.5, lambda = 0.1013 )
#>                                                                lambda_search
#> 1 nlambda = 100, lambda.max = 10.132, lambda.min = 0.1013, lambda.1se = -1.0
#>   number_of_predictors_total number_of_active_predictors
#> 1                          2                           2
#>   number_of_iterations                                training_frame
#> 1                  100 frame_rdd_31_ad5c4e88ec97eb8ccedae9475ad34e02
#>
#> Coefficients: glm coefficients
#>       names coefficients standardized_coefficients
#> 1 Intercept    38.941654                 20.090625
#> 2       cyl    -1.468783                 -2.623132
#> 3        wt    -3.034558                 -2.969186
#>
#> H2ORegressionMetrics: glm
#> ** Reported on training data. **
#>
#> MSE:  6.017684
#> RMSE:  2.453097
#> MAE:  1.940985
#> RMSLE:  0.1114801
#> Mean Residual Deviance :  6.017684
#> R^2 :  0.8289895
#> Null Deviance :1126.047
#> Null D.o.F. :31
#> Residual Deviance :192.5659
#> Residual D.o.F. :29
#> AIC :156.2425
```

```{r eval=FALSE}
spark_disconnect(sc)
```

## Connecting through Livy

[Livy](https://github.com/cloudera/livy) enables remote connections to Apache 
Spark clusters. However, please notice that connecting to Spark clusters through 
Livy is much slower than any other connection method.

Before connecting to Livy, you will need the connection information to an 
existing service running Livy. Otherwise, to test `livy` in your local 
environment, you can install it and run it locally as follows:

```{r eval=FALSE}
livy_install()
```

```{r livy-start, eval = FALSE}
livy_service_start()
```

To connect, use the Livy service address as `master` and `method = "livy"` in 
`spark_connect()`. Once connection completes, use `sparklyr` as usual, for instance:

```{r livy-connect, eval = FALSE}
sc <- spark_connect(master = "http://localhost:8998", method = "livy", version = "3.0.0")
copy_to(sc, iris, overwrite = TRUE)

spark_disconnect(sc)
```

Once you are done using `livy` locally, you should stop this service with:

```{r livy-stop, eval = FALSE}
livy_service_stop()
```

To connect to remote `livy` clusters that support basic authentication connect as:

```{r eval=FALSE}
config <- livy_config(username="<username>", password="<password>")
sc <- spark_connect(master = "<address>", method = "livy", config = config)
spark_disconnect(sc)
```

## Connecting through Databricks Connect v2

`sparklyr` is able to interact with [Databricks Connect v2](https://docs.databricks.com/en/dev-tools/databricks-connect/index.html) 
via a new extension called `pysparklyr`. To learn how to use, and the latest 
updates on this integration see 
[the article in `sparklyr`'s official website](https://spark.posit.co/deployment/databricks-connect.html).