A previous article covered sums and averages on the whole data set. In this article, we show how to use Collectors.groupingBy() to perform SQL-like grouping on tabular data.

Define a POJO

Here is the POJO that we use in the examples below. It represents a baseball player.

public class Player {
    private int year;
    private String teamID;
    private String lgID;
    private String playerID;
    private int salary;


}

We load the following sample data representing player salaries.

yearID,teamID,lgID,playerID,salary
1985,ATL,NL,barkele01,870000
1985,ATL,NL,bedrost01,550000
...

Load CSV Data

Since the CSV is quite simple (no quoted fields, no commas inside fields, etc. See this article for code covering those cases), we use a simple regex-based CSV parser. The data is loaded into a list of POJOs.

Pattern pattern = Pattern.compile(",");
try (BufferedReader in = new BufferedReader(new FileReader(filename));){
    List<Player> players = in
        .lines()
        .skip(1)
        .map(line -> {
                String[] arr = pattern.split(line);
                return new Player(Integer.parseInt(arr[0]),
                                  arr[1],
                                  arr[2],
                                  arr[3],
                                  Integer.parseInt(arr[4]));
            })
        .collect(Collectors.toList());
}

Group by a Single Field

In an illustration of a simple usage of groupingBy(), we show how to group the data by year. The result is a map of year and players list grouped by the year.

    Map<Integer,List<Player>> grouped = in
        .lines()
        .skip(1)
        .map(line -> {
                String[] arr = pattern.split(line);
                return new Player(Integer.parseInt(arr[0]),
                                  arr[1],
                                  arr[2],
                                  arr[3],
                                  Integer.parseInt(arr[4]));
            })
        .collect(Collectors.groupingBy(x->x.getYear()));
    grouped
        .entrySet()
        .stream()
        .forEach(System.out::println);


1985=[{1985,ATL,NL,barkele01,870000}, {1985,ATL,NL ...
1986=[{1986,ATL,NL,ackerji01,367500}, {1986,ATL,NL ...

By Multiple Fields

Grouping by multiple fields is a little bit more involved because the resulting map is keyed by the list of fields selected. We create a List in the groupingBy() clause to serve as the key of the map. The value is the Player object.

The output map is printed using the for-each block below.

Map<List<String>,List<Player>> grouped = in
    .lines()
    .skip(1)
    .map(line -> {
            String[] arr = pattern.split(line);
            return new Player(Integer.parseInt(arr[0]),
                              arr[1],
                              arr[2],
                              arr[3],
                              Integer.parseInt(arr[4]));
        })
    .collect(Collectors.groupingBy(x -> {
                return new ArrayList<String>(Arrays.asList(Integer.toString(x.getYear()), x.getTeamID()));
            }));
grouped
    .entrySet()
    .stream()
    .forEach(x -> {
            System.out.println(x.getKey());
            x.getValue().stream()
                .forEach(p -> System.out.printf(" ( %2s %-10s %-10d )%n", p.getLgID(), p.getPlayerID(), p.getSalary()));
        });


[1987, PHI]
 ( NL aguaylu01 325000 )
 ( NL bedrost01 1050000 )
 ( NL calhoje01 85000 )
...

Eliminating the intermediate variable grouped, we have the entire processing pipeline in a single statement as shown below.

in
    .lines()
    .skip(1)
    .map(line -> {
            String[] arr = pattern.split(line);
            return new Player(Integer.parseInt(arr[0]),
                              arr[1],
                              arr[2],
                              arr[3],
                              Integer.parseInt(arr[4]));
        })
    .collect(Collectors.groupingBy(x -> {
                return new ArrayList<String>(Arrays.asList(Integer.toString(x.getYear()), x.getTeamID()));
            }))
    .entrySet()
    .stream()
    .forEach(x -> {
            System.out.println(x.getKey());
            x.getValue().stream()
                .forEach(p -> System.out.printf(" ( %2s %-10s %-10d )%n", p.getLgID(), p.getPlayerID(), p.getSalary()));
        });

Collecting Into a Set

The code above collects the results into the following type:

Map<List<String>,List<Player>>

Instead of storing the values in a List (inside the Map), how can we put it into a Set? We use the groupedBy() version that accepts the value-type.

Map<List<String>,Set<Player>> grouped = in
    .lines()
    .skip(1)
    .map(line -> {
            String[] arr = pattern.split(line);
            return new Player(Integer.parseInt(arr[0]),
                              arr[1],
                              arr[2],
                              arr[3],
                              Integer.parseInt(arr[4]));
        })
    .collect(Collectors.groupingBy(x -> return new ArrayList<String>(Arrays.asList(x.getTeamID(), x.getLgID())), Collectors.toSet()));

The first argument to groupingBy() is a lambda function which accepts a Player and returns a List of fields to group-by. The second argument is lambda which creates the Collection to place the results of the group-by.

This invocation collects the values into a Set which has no ordering. To retrieve values in the same order in which they were placed into the Set, use a LinkedHashSet as shown below (only the relevant portion is shown):

...
.collect(Collectors.groupingBy(x -> return new ArrayList<String>(Arrays.asList(x.getTeamID(), x.getLgID())), Collectors.toCollection(LinkedHashSet::new)));

Summarizing Field Values

Forget about collecting the results into a List or Set or whatever. What if we want to compute an aggregate of a field value? Say, for example, the sum of all salaries paid to players grouped by team and league.  Simple! Use a summingInt() as the second argument to groupingBy(). Remember the second argument collects the value-type of the group into the map.

Map<List<String>,Integer> grouped = in
    .lines()
    .skip(1)
    .map(line -> {
            String[] arr = pattern.split(line);
            return new Player(Integer.parseInt(arr[0]),
                              arr[1],
                              arr[2],
                              arr[3],
                              Integer.parseInt(arr[4]));
        })
    .collect(Collectors.groupingBy(x -> new ArrayList<String>(Arrays.asList(x.getTeamID(), x.getLgID())), Collectors.summingInt(Player::getSalary)));

Summary

Summarizing using grouping by is a useful technique in data analysis. Normally these are available in SQL databases. However using the Java 8 Streams and Collections facility, it is possible to use these techniques on Java collections. Some examples included grouping and summarizing with aggregate operations.