How to get dplyr::summarize_all to work on a sparkdataframe using databricks?

user_8377495 0
Viewed 611

Question

I have a massive Spark Dataframe, called x. I am using databricks. x is billions of records long, too large to collect onto a single machine. What do I have to do to get this to work?:

dplyr::summarize_all(x,mean)

More Info

This is the error message I currently get:

Error in UseMethod("tbl_vars") : 
  no applicable method for 'tbl_vars' applied to an object of class "SparkDataFrame"

and

class(x)

returns: [1] "SparkDataFrame" attr(,"package") [1] "SparkR"

The book, Mastering Spark with R , has an example of loading up a tiny r data frame, and running summarize_all on it:

cars <- copy_to(sc, mtcars)
summarize_all(cars, mean)

Note the above code works on my databricks cluster and returns a nice block of text:

# Source: spark<?> [?? x 11]
    mpg   cyl  disp    hp  drat    wt  qsec    vs    am  gear  carb
  <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1  20.1  6.19  231.  147.  3.60  3.22  17.8 0.438 0.406  3.69  2.81

The same book leads me to believe I can use this and similar functions on huge spark dataframes.

and also

class(cars)

returns:

[1] "tbl_spark" "tbl_sql"   "tbl_lazy"  "tbl"

It seems obvious that I need to convert my spark dataframe to a tbl_spark, tbl_sql, tbl_lazy or tbl so that I can pass it to dplyr::summarize_all, but I have searched all over the place and asked experts and cannot figure out how to do this.

dplyr databricks apache-spark r
1 Answers

You're right that SparkR and sparklyr are different APIs that don't play well together. You can convert the SparkR data frame to be used with sparklyr by using a temp table.

Here's an example SparkR data frame.

sc <- sparklyr::spark_connect(method = "databricks")

x_sparkr <- SparkR::sql("SELECT 1 AS a UNION SELECT 2")

Create the temp table.

SparkR::registerTempTable(x_sparkr, "temp_x")

Load it into sparklyr.

x_sparklyr <- dplyr::tbl(sc, "temp_x")

dplyr::summarize_all(x_sparklyr, mean)
#> # Source: spark<?> [?? x 1]
#>       a
#>   <dbl>
#> 1   1.5
user_17519610
Related