Is there a Java equivalent of Python's defaultdict?

Question

In Python, the defaultdict class provides a convenient way to create a mapping from key -> [list of values], in the following example,

from collections import defaultdict
d = defaultdict(list)
d[1].append(2)
d[1].append(3)
# d is now {1: [2, 3]}

Is there an equivalent to this in Java?

Solution 1

There is nothing that gives the behaviour of default dict out of the box. However creating your own default dict in Java would not be that difficult.

import java.util.ArrayList;
import java.util.HashMap;
import java.util.List;

public class DefaultDict<K, V> extends HashMap<K, V> {

    Class<V> klass;
    public DefaultDict(Class klass) {
        this.klass = klass;    
    }

    @Override
    public V get(Object key) {
        V returnValue = super.get(key);
        if (returnValue == null) {
            try {
                returnValue = klass.newInstance();
            } catch (Exception e) {
                throw new RuntimeException(e);
            }
            this.put((K) key, returnValue);
        }
        return returnValue;
    }    
}

This class could be used like below:

public static void main(String[] args) {
    DefaultDict<Integer, List<Integer>> dict =
        new DefaultDict<Integer, List<Integer>>(ArrayList.class);
    dict.get(1).add(2);
    dict.get(1).add(3);
    System.out.println(dict);
}

This code would print: {1=[2, 3]}

Solution 2

In most common cases where you want a defaultdict, you'll be even happier with a properly designed Multimap or Multiset, which is what you're really looking for. A Multimap is a key -> collection mapping (default is an empty collection) and a Multiset is a key -> int mapping (default is zero).

Guava provides very nice implementations of both Multimaps and Multisets which will cover almost all use cases.

But (and this is why I posted a new answer) with Java 8 you can now replicate the remaining use cases of defaultdict with any existing Map.

• getOrDefault(), as the name suggests, returns the value if present, or returns a default value. This does not store the default value in the map.
• computeIfAbsent() computes a value from the provided function (which could always return the same default value) and does store the computed value in the map before returning.

If you want to encapsulate these calls you can use Guava's ForwardingMap:

public class DefaultMap<K, V> extends ForwardingMap<K, V> {
  private final Map<K, V> delegate;
  private final Supplier<V> defaultSupplier;

  /**
   * Creates a map which uses the given value as the default for <i>all</i>
   * keys. You should only use immutable values as a shared default key.
   * Prefer {@link #create(Supplier)} to construct a new instance for each key.
   */
  public static DefaultMap<K, V> create(V defaultValue) {
    return create(() -> defaultValue);
  }

  public static DefaultMap<K, V> create(Supplier<V> defaultSupplier) {
    return new DefaultMap<>(new HashMap<>(), defaultSupplier);
  }

  public DefaultMap<K, V>(Map<K, V> delegate, Supplier<V> defaultSupplier) {
    this.delegate = Objects.requireNonNull(delegate);
    this.defaultSupplier = Objects.requireNonNull(defaultSupplier);
  }

  @Override
  public V get(K key) {
    return delegate().computeIfAbsent(key, k -> defaultSupplier.get());
  }
}

Then construct your default map like so:

Map<String, List<String>> defaultMap = DefaultMap.create(ArrayList::new);

Solution 3

in addition to apache collections, check also google collections:

A collection similar to a Map, but which may associate multiple values with a single key. If you call put(K, V) twice, with the same key but different values, the multimap contains mappings from the key to both values.

Solution 4

You can use MultiMap from Apache Commons.

Solution 5

In Java 8+ you can use:

map.computeIfAbsent(1, k -> new ArrayList<Integer>()).add(2);

View more solutions

Comments

[Mark Byers]

Link: commons.apache.org/collections/api/org/apache/commons/…

[Lambda Fairy]

Instead of using a Class, you can also try passing a Guava Supplier -- see docs.guava-libraries.googlecode.com/git-history/v10.0/javado‌​c/…

[Soulman]

Or, if you don't want the Guava dependeny, just define your own Supplier<V> interface in DefaultDict.

[sam boosalis]

i would prefer constructing the DefaultDict with my own value: public DefaultDict(V value) { this.value = value; }

[Ar5hv1r]

It would be more usable if instead of extending from HashMap you used ForwardingMap and allowed the caller to specify the backing map. Prefer composition to inheritance.

[antak]

I didn't downvote. Good idea all in all. However I don't think you can name your variable default, and I think create(V) is just asking users to shoot themselves in their foot.

[Ar5hv1r]

Thanks, you're absolutely right. I agree create(V) is a little risky, but I suspect it'd be missed if it was absent. I've added a comment, at least, to call out the risk.

[dtk]

The MultiMap is Deprecated. since 4.1, use MultiValuedMap instead. commons.apache.org/proper/commons-collections/apidocs/org/…

[user102008]

Wouldn't this behave incorrectly if you stored a value of null for a certain key into the map? When you call get(), it would behave as if the key didn't exist (even though it does exist, as determined by containsKey()), and they create a new value and overwrite the existing value in the map.

[rocksNwaves]

How would this work if your default value type was something other than a collection? For instance an Integer with value of 1?

[rocksNwaves]

Also, it seems that this results in an unsafe type cast that cannot be made safe because of type erasure. I'd roll with it, but my boss certainly wouldn't.