Add and Search Word - Data structure design
Design a data structure that supports the following two operations:
void addWord(word)
bool search(word)
search(word) can search a literal word or a regular expression string containing only letters a-z or .. A . means it can represent any one letter.
For example:
addWord("bad")
addWord("dad")
addWord("mad")
search("pad") -> false
search("bad") -> true
search(".ad") -> true
search("b..") -> true
Note:
You may assume that all words are consist of lowercase letters a-z.
You should be familiar with how a Trie works. If not, please work on this problem: Implement Trie (Prefix Tree) first.
Tips:
O(n)
先建立trie tree, 然后用BFS搜索。需要注意的是
- trie树中isKey代表是不是单词的结束。
- 碰到‘.‘要把当前层不是null的全部加入queue再搜索。
Code:
public class WordDictionary {
class TrieNode {
boolean isKey;
TrieNode[] children;
public TrieNode() {
children = new TrieNode[26];
for (int i = 0; i < 26; i++) {
children[i] = null;
}
}
}
TrieNode root = new TrieNode();
// Adds a word into the data structure.
public void addWord(String word) {
TrieNode copy = root;
for (char c : word.toCharArray()) {
if (copy.children[c - 'a'] == null) {
copy.children[c - 'a'] = new TrieNode();
}
copy = copy.children[c - 'a'];
}
copy.isKey = true;
}
// Returns if the word is in the data structure. A word could
// contain the dot character '.' to represent any one letter.
public boolean search(String word) {
return dfs(word, 0, root);
}
private boolean dfs(String word, int depth, TrieNode node) {
if (depth == word.length()) {
return node.isKey;
}
char c = word.charAt(depth);
if (c != '.') {
return node.children[c - 'a'] != null && dfs(word, depth + 1, node.children[c - 'a']);
} else {
for (TrieNode child : node.children) {
if (child != null) {
if (dfs(word, depth + 1, child)) {
return true;
}
}
}
return false;
}
}
}
bfs
public class WordDictionary {
class TrieNode {
boolean isKey;
TrieNode[] children;
public TrieNode() {
isKey = false;
children = new TrieNode[26];
for (int i = 0; i < 26; i++) {
children[i] = null;
}
}
}
private TrieNode root;
public WordDictionary() {
root = new TrieNode();
}
// Adds a word into the data structure.
public void addWord(String word) {
TrieNode copy = root;
for (int i = 0; i < word.length(); i++) {
if (copy.children[word.charAt(i) - 'a'] == null) {
copy.children[word.charAt(i) - 'a'] = new TrieNode();
copy = copy.children[word.charAt(i) - 'a'];
}
else copy = copy.children[word.charAt(i) - 'a'];
}
copy.isKey = true;
}
// Returns if the word is in the data structure. A word could
// contain the dot character '.' to represent any one letter.
public boolean search(String word) {
Queue<TrieNode> queue = new LinkedList<>();
queue.offer(root);
for (int i = 0; i < word.length(); i++) {
if (word.charAt(i) == '.') {
int size = queue.size();
if (size == 0) {
return false;
}
for (int j = 0; j < size; j++) {
TrieNode current = queue.poll();
for (int x = 0; x < 26; x++) {
if (current.children[x] != null) {
queue.offer(current.children[x]);
}
}
}
}
else {
int size = queue.size();
if (size == 0) {
return false;
}
for (int j = 0; j < size; j++) {
TrieNode current = queue.poll();
if (current.children[word.charAt(i) - 'a'] != null) {
queue.offer(current.children[word.charAt(i) - 'a']);
}
}
}
}
int size = queue.size();
for (int i = 0; i < size; i++) {
TrieNode current = queue.poll();
if (current.isKey) {
return true;
}
}
return false;
}
}