Module org.jgrapht.core
Package org.jgrapht.alg.similarity

Class ZhangShashaTreeEditDistance<V,E>

java.lang.Object
org.jgrapht.alg.similarity.ZhangShashaTreeEditDistance<V,E>
Type Parameters:
V - graph vertex type
E - graph edge type
public class ZhangShashaTreeEditDistance<V,E> extends Object
Dynamic programming algorithm for computing edit distance between trees.

The algorithm is originally described in Zhang, Kaizhong & Shasha, Dennis. (1989). Simple Fast Algorithms for the Editing Distance Between Trees and Related Problems. SIAM J. Comput.. 18. 1245-1262. 10.1137/0218082.

The time complexity of the algorithm is $O(|T_1|\cdot|T_2|\cdot min(depth(T_1),leaves(T_1)) \cdot min(depth(T_2),leaves(T_2)))$. Space complexity is $O(|T_1|\cdot |T_2|)$, where $|T_1|$ and $|T_2|$ denote number of vertices in trees $T_1$ and $T_2$ correspondingly, $leaves()$ function returns number of leaf vertices in a tree.

The tree edit distance problem is defined in a following way. Consider $2$ trees $T_1$ and $T_2$ with root vertices $r_1$ and $r_2$ correspondingly. For those trees there are 3 elementary modification actions:

  • Remove a vertex $v$ from $T_1$.
  • Insert a vertex $v$ into $T_2$.
  • Change vertex $v_1$ in $T_1$ to vertex $v_2$ in $T_2$.
The algorithm assigns a cost to each of those operations which also depends on the vertices. The problem is then to compute a sequence of edit operations which transforms $T_1$ into $T_2$ and has a minimum cost over all such sequences. Here the cost of a sequence of edit operations is defined as sum of costs of individual operations.

The algorithm is based on a dynamic programming principle and assigns a label to each vertex in the trees which is equal to its index in post-oder traversal. It also uses a notion of a keyroot which is defined as a vertex in a tree which has a left sibling. Additionally a special $l()$ function is introduced with returns for every vertex the index of its leftmost child wrt the post-order traversal in the tree.

Solving the tree edit problem distance is divided into computing edit distance for every pair of subtrees rooted at vertices $v_1$ and $v_2$ where $v_1$ is a keyroot in the first tree and $v_2$ is a keyroot in the second tree.

Author:
Semen Chudakov

  • Constructor Details

    • ZhangShashaTreeEditDistance

      public ZhangShashaTreeEditDistance(Graph<V,E> tree1, V root1, Graph<V,E> tree2, V root2)
      Constructs an instance of the algorithm for the given tree1, root1, tree2 and root2. This constructor sets following default values for the distance functions. The insertCost and removeCost always return $1.0$, the changeCost return $0.0$ if vertices are equal and 1.0 otherwise.
      Parameters:
      tree1 - a tree
      root1 - root vertex of tree1
      tree2 - a tree
      root2 - root vertex of tree2

    • ZhangShashaTreeEditDistance

      public ZhangShashaTreeEditDistance(Graph<V,E> tree1, V root1, Graph<V,E> tree2, V root2, ToDoubleFunction<V> insertCost, ToDoubleFunction<V> removeCost, ToDoubleBiFunction<V,V> changeCost)
      Constructs an instance of the algorithm for the given tree1, root1, tree2, root2, insertCost, removeCost and changeCost.
      Parameters:
      tree1 - a tree
      root1 - root vertex of tree1
      tree2 - a tree
      root2 - root vertex of tree2
      insertCost - cost function for inserting a node into tree1
      removeCost - cost function for removing a node from tree2
      changeCost - cost function of changing a node in tree1 to a node in tree2

  • Method Details

    • getDistance

      public double getDistance()
      Computes edit distance for tree1 and tree2.
      Returns:
      edit distance between tree1 and tree2

    • getEditOperationLists

      public List<ZhangShashaTreeEditDistance.EditOperation<V>> getEditOperationLists()
      Computes a list of edit operations which transform tree1 into tree2.
      Returns:
      list of edit operations