Home

Awesome

Build status [NuGet Package] [Get Commercial License]

DawgSharp, a clever string dictionary in C#

DAWG (Directed Acyclic Word Graph) is a data structure for storing and searching large word lists and dictionaries. It can be 40x more efficient than the .NET Dictionary class for certain types of data.

As an example, my website hosts a 2 million word dictionary which used to take up 56 meg on disk and took 7 seconds to load (when using Dictionary and BinarySerializer). After switching to DAWG, it now takes 1.4 meg on disk and 0.3 seconds to load.

How is this possible? Why is the standard Dictionary not as clever as DAWG? The thing is, DAWG works well with natural language strings and may not work as well for generated strings such as license keys (OIN1r4Be2su+UXSeOj0TaQ). Human language words tend to have lots of common letter sequences eg -ility in ability, possibility, agility etc and the algorithm takes advantage of that by finding those sequences and storing them only once for multiple words. DAWG has also proved useful in representing DNA data (sequences of genes). The history of DAWG dates back as far as 1985. For more backgroud, google DAWG or DAFSA (Deterministic Acyclic Finite State Automaton).

DawgSharp is an implementation of DAWG, one of many. What makes it special?

Usage

In this example we will simulate a usage scenario involving two programs, one to generate the dictionary and write it to disk and the other to load that file and use the read-only dictionary for lookups.

First get the code by cloning this repository or installing the NuGet package.

Create and populate a DawgBuilder object:

var words = new [] { "Aaron", "abacus", "abashed" };

var dawgBuilder = new DawgBuilder <bool> (); // <bool> is the value type.
                                             // Key type is always string.
foreach (string key in words)
{
    dawgBuilder.Insert (key, true);
}

(Alternatively, do var dawgBuilder = words.ToDawgBuilder(key => key, _ => true);)

Call BuildDawg on it to get the compressed version and save it to disk:

Dawg<bool> dawg = dawgBuilder.BuildDawg (); 
// Computer is working.  Please wait ...

using (var file = File.Create ("DAWG.bin")) 
    dawg.SaveTo (file);

Now read the file back in and check if a particular word is in the dictionary:

var dawg = Dawg <bool>.Load (File.Open ("DAWG.bin"));

if (dawg ["chihuahua"])
{
    Console.WriteLine ("Word is found.");
}

The Value Type, <TPayload>

The Dawg and DawgBuilder classes take a template parameter called <TPayload>. It can be any type you want. Just to be able to test if a word is in the dictionary, a bool is enough. You can also make it an int or a string or a custom class. But beware of one important limitation. DAWG works well only when the set of values that TPayload can take is relatively small. The smaller the better. Eg if you add a definition for each word, it will make each entry unique and your graph will become a tree (which may not be too bad!).

MatchPrefix()

One other attractive side of DAWG is its ability to efficiently retrieve all words starting with a particular substring:

dawg.MatchPrefix("awe")

The above query will return an IEnumerable<KeyValuePair> which might contain keys such as awe, aweful and awesome. The call dawg.MatchPrefix("") will return all items in the dictionary.

If you need to look up by suffix instead, there is no MatchSuffix method. But the desired effect can be achieved by adding the reversed keys and then using MatchPrefix() on the reversed keys:

dawgBuilder.Insert("ability".Reverse(), true);
...
dawg.MatchPrefix("ility".Reverse())

GetPrefixes()

GetPrefixes() returns all dictionary items whose keys are substrings of a given string. For example:

dawg.GetPrefixes("awesomenesses")

Might return keys such as awe, awesome, awesomeness and finally awesomenesses.

GetLongestCommonPrefixLength()

One other neat feature is the method int GetLongestCommonPrefixLength(IEnumerable<char> word). If word is found in the dictionary, it will return its length; if not, it will return the length of the longest word that is found in the dictionary and that is also the beginning of the given word. For example, if prepare is in the dictionary but preempt is not, then dawg.GetLongestCommonPrefixLength("preempt") will return 3 which is the length of "pre".

Thread Safety

The DawgBuilder class is not thread-safe and must be accessed by only one thread at any particular time.

The Dawg class is immutable and thus thread-safe.

MultiDawg

The MultiDawg class can store multiple values agaist a single string key in a very memory-efficient manner.

Future plans

More usage scenarios

The API was designed to fit a particular usage scenario (see above) and can be extended to support other scenarios eg being able to add new words to the dictionary after it's been compacted. I just didn't need this so it's not implemented. You won't get any exceptions. There is just no Insert method on the Dawg class.

Better API

Implement the IDictionary interface on both DawgBuilder and Dawg (#5).

Literature

Competing Implementations

License

DawgSharp is licensed under GPLv3 which means it can be used free of charge in open-sources projects. Read the full license

If you would like to use DawgSharp in a proprietary project, please purchase a commercial license at http://morpher.co.uk.