Trie

15 min read

What is a Trie?

A Trie, also known as a prefix tree or digital tree, is a tree-like data structure that is used to efficiently store and retrieve keys in a dataset of strings. The Trie is especially useful when dealing with dynamic sets of strings and allows for fast searching, insertion, and deletion of keys.

Components of a Trie

  • Nodes: Each node represents a single character of a string or a key. Each node has a set of children, which correspond to the next characters in the keys.
  • Root: The root node represents the start of the Trie and usually doesn’t hold any character value. It’s associated with an empty string. Each path from the root to a leaf represents a key in the Trie.
  • Edges: Edges connect nodes and represent the next character in the sequence.
  • End of Word Marker: A special marker or flag that indicates the end of a valid word. Each node has the end of word marker as a boolean flag.

Examples

The following Trie stores words: bag, ball, bar, bark & bat

grllobaorktt

Operations on a Trie (with code examples)

Creation

Creation of an empty Trie with root node representing empty string.

go 
 1package main
 2
 3import "fmt"
 4
 5type TrieNode struct {
 6	children  map[rune]*TrieNode
 7	endOfWord bool
 8}
 9
10type Trie struct {
11	root *TrieNode
12}
13
14func NewTrie() *Trie {
15	return &Trie{root: &TrieNode{children: make(map[rune]*TrieNode)}}
16}
17
18func main() {
19  trie := NewTrie()
20}

Insertion

Adding a word involves creating new nodes for each character if they don’t already exist and marking the last character node as the end of the word.

go 
 1func (t *Trie) Insert(word string) {
 2	node := t.root
 3	for _, char := range word {
 4		if _, exists := node.children[char]; !exists {
 5			node.children[char] = &TrieNode{children: make(map[rune]*TrieNode)}
 6		}
 7		node = node.children[char]
 8	}
 9	node.endOfWord = true
10}
11
12func main() {
13  ...
14  wordsToInsert := []string{"bat", "ball", "bark", "belt", "better", "bet", "betting", "car", "cart", "call", "called"}
15
16  // Inserting words into the Trie
17  for _, word := range wordsToInsert {
18    trie.Insert(word)
19  }
20}
How the above code works:
  • Set the current node pointer to root node (empty string).
  • For each character in the given word “bat” check if it’s one of the children.
  • Starting with ‘b’. Since the trie is empty, the current node (root) has no children, hence we create a new node for ‘b’ and set it as one of the children of current node (root).
  • Set the current node pointer to the child (‘b’) node.
  • Moving on to ‘a’. Since current node (‘b’) has no children, we create a new node for ‘a’ and set it as one of the children of current node (‘b’).
  • Set the current node pointer to the child (‘a’) node.
  • Moving on to ’t’. Since current node (‘a’) has no children, we create a new node for ’t’ and set it as one of the children of current node (‘a’).
  • Set the current node pointer to the child (’t’) node.
  • Since all characters of the word are finished, we set the current node’s (’t’) endOfWord flag to true as it’s a complete path for the word “bat” in the trie.
  • Now the trie is no longer empty for the next word to be inserted.
  • Set the current node pointer to root node again.
  • For each character in the given word “ball” check if it’s one of the children.
  • Starting with ‘b’. The current node (root) already has ‘b’ node as one of it’s children. We skip creating a new node.
  • Set the current node pointer to the child (‘b’) node.
  • Moving on to ‘a’. The current node (‘b’) already has ‘a’ node as one of it’s children. We skip creating a new node.
  • Set the current node pointer to the child (‘a’) node.
  • Moving on to ’l’. Since current node (‘a’) doesn’t have ’l’ node as a child, we create a new node for ’l’ and set it as one of the children of current node (‘a’).
  • Set the current node pointer to the child (’l’) node.
  • Moving on to ’l’. Since current node (’l’) has no children, we create a new node for 2nd ’l’ and set it as one of the children of current node (1st ’l’).
  • Set the current node pointer to the child (2nd ’l’) node.
  • Since all characters of the word are finished we set the current node’s (2nd ’l’) endOfWord flag to true as it’s a complete path for the word “ball” in the trie.
  • Same steps are repeated until all words are inserted.

The diagram below repsents the state of the trie after all words (“bat”, “ball”, “bark”, “belt”, “better”, “bet”, “betting”, “car”, “cart”, “call”, “called”) are inserted. Here minus (-) sign next to some characters represent an end of word.

t-all-rk-blt-eerr-otto-ting-rt--called--

To search for a word, traverse the Trie following the sequence of characters in the word. If you reach the end and the endOfWord flag is present, the word exists in the Trie.

go 
 1func (t *Trie) Search(word string) bool {
 2	node := t.root
 3	for _, char := range word {
 4		if _, exists := node.children[char]; !exists {
 5			return false
 6		}
 7		node = node.children[char]
 8	}
 9	return node.endOfWord
10}
11
12func main() {
13  ...
14  fmt.Println("Search Results:")
15  for _, word := range wordsToInsert {
16		fmt.Printf("Word '%s' found?: %v\n", word, trie.Search(word))
17	}
18}

Output:

shell 
 1Search Results Before Deletion:
 2Word 'bat' found?: true
 3Word 'ball' found?: true
 4Word 'bark' found?: true
 5Word 'belt' found?: true
 6Word 'better' found?: true
 7Word 'bet' found?: true
 8Word 'betting' found?: true
 9Word 'car' found?: true
10Word 'cart' found?: true
11Word 'call' found?: true
12Word 'called' found?: true
How the above code works:
  • Let’s assume we are searching for the word “bat”.
  • Set the current node pointer to the root node.
  • Starting with ‘b’. Since ‘b’ is one of the children of current node (root). We set current node as node ‘b’.
  • Moving on to ‘a’. Since ‘a’ is one of the children of current node (‘b’). We set the current node as node ‘a’.
  • Moving on to ’t’. Since ’t’ is one of the children of current node (‘a’). We set the current node as node ’t’.
  • We have compared all the characters.
  • Since the the node ’t’ has endOfWord flag set to true, it means that the word path points to a complete word i.e. “bat”. Hence the search result returns true.
  • Now, let’s assume we are searching for the word “best”.
  • Set the current node pointer to the root node.
  • Starting with ‘b’. Since ‘b’ is one of the children of current node (root). We set current node as node ‘b’.
  • Moving on to ’e’. Since ’e’ is one of the children of current node (‘b’). We set the current node as node ‘a’.
  • Moving on to ’s’. Since ’s’ is not one of the children of current node (’e’). We return false as the word is not present in the Trie.
  • Let’s assume we inserted the word “better” but didn’t insert “bet”. Note that “better” has “bet” as prefix. However since the endOfWord flag for the node ’t’ is false so the word “bet” is not found.

Useful for auto-complete systems. The code below checks if the words matching the given prefix exist in a Trie.

go 
 1func (t *Trie) StartsWith(prefix string) bool {
 2	node := t.root
 3	for _, char := range prefix {
 4		if _, exists := node.children[char]; !exists {
 5			return false
 6		}
 7		node = node.children[char]
 8	}
 9	return true
10}
11
12func main() {
13  ...
14  prefix := "ba"
15  fmt.Printf("Words starting with '%s' found?: %v\n", prefix, trie.StartsWith(prefix))
16}

Output:

shell 
1Words starting with 'ba' found?: true
How the above code works:
  • Code follows the same strategy as the search however it doesn’t check for the endOfWord flag since we are just checking for prefix.
  • If all characters in the prefix are found in the successive children nodes, the for loop ends successfully and true is returned at the end.

Autocomplete

Find all words with a given prefix by traversing the Trie to the end of the prefix and then listing all descendants.

go 
 1func (t *Trie) GetWordsWithPrefix(prefix string) []string {
 2	node := t.root
 3	for _, char := range prefix {
 4		if _, exists := node.children[char]; !exists {
 5			return nil // No words with the given prefix
 6		}
 7		node = node.children[char]
 8	}
 9	return t.collectWords(node, prefix)
10}
11
12func (t *Trie) collectWords(node *TrieNode, prefix string) []string {
13	var words []string
14
15	if node.endOfWord {
16		words = append(words, prefix)
17	}
18
19	for char, childNode := range node.children {
20		words = append(words, t.collectWords(childNode, prefix+string(char))...)
21	}
22
23	return words
24}
25
26func main() {
27  ...
28	prefix = "ba"
29	fmt.Printf("Words with prefix '%s':\n", prefix)
30	matchingWords := trie.GetWordsWithPrefix(prefix)
31	for _, word := range matchingWords {
32		fmt.Println(word)
33	}
34
35	prefix = "ca"
36	fmt.Printf("\nWords with prefix '%s':\n", prefix)
37	matchingWords = trie.GetWordsWithPrefix(prefix)
38	for _, word := range matchingWords {
39		fmt.Println(word)
40	}
41}

Output:

shell 
 1Words with prefix 'ba':
 2bat
 3ball
 4bark
 5
 6Words with prefix 'ca':
 7car
 8cart
 9call
10called
How the above code works:
  • Let’s assume we are searching for prefix “ba”.
  • The GetWordsWithPrefix function will start at the root and traverse through nodes for ‘b’ and ‘a’.
  • After this traversal, the node will be pointing to the node representing ‘a’ in the Trie.
t-roballo-trk-
  • The collectWords function will be called starting from the node representing ‘a’ with the prefix “ba”.
  • The function will recursively explore nodes for ‘t’, ‘l’, and ‘r’ (the children of ‘a’).
  • Going down ’t’ the endOfWord flag is found to be true, the word “bat” is collected.
  • Going down ’l’ the endOfWord flag is found to be false, we iterate over children of ’l’
  • Going down 2nd ’l’ the endOfWord flag is found to be true, the word “ball” is collected.
  • Coming one level back up and going down ‘r’ the endOfWord flag is found to be false, we iterate over children of ‘r’
  • Going down to ‘k’ the endOfWord flag is found to be true, the word “bark” is collected.
  • Going back up to the root all the collections are merged to the final result: [“bat”, “ball”, “bark”].

Deletion

Deleting a word involves removing the end of word marker and pruning the Trie if the nodes are not part of another word.

go 
 1func (t *Trie) Delete(word string) bool {
 2	return t.deleteHelper(t.root, word, 0)
 3}
 4
 5func (t *Trie) deleteHelper(node *TrieNode, word string, depth int) bool {
 6	if node == nil {
 7		return false
 8	}
 9
10	// If we've reached the end of the word
11	if depth == len(word) {
12		// This word doesn't exist
13		if !node.endOfWord {
14			return false
15		}
16
17		// Unmark the end of word node
18		node.endOfWord = false
19
20		// If this node has no children, it can be deleted
21		return len(node.children) == 0
22	}
23
24	char := rune(word[depth])
25	if t.deleteHelper(node.children[char], word, depth+1) {
26		// Remove the child node reference if it's no longer needed
27		delete(node.children, char)
28
29		// Return true if the current node is no longer a part of another word
30		return !node.endOfWord && len(node.children) == 0
31	}
32
33	return false
34}
35
36func main() {
37  ...
38  wordsToDelete := []string{"bet", "belt", "call"}
39
40	// Deleting words from the Trie
41	fmt.Println("\nDeleting words:", wordsToDelete)
42	for _, word := range wordsToDelete {
43		trie.Delete(word)
44	}
45
46	fmt.Println("\nSearch Results After Deletion:")
47	for _, word := range wordsToInsert {
48		fmt.Printf("Word '%s' found?: %v\n", word, trie.Search(word))
49	}
50}

Output:

shell 
 1Deleting words: [bet belt call]
 2
 3Search Results After Deletion:
 4Word 'bat' found?: true
 5Word 'ball' found?: true
 6Word 'bark' found?: true
 7Word 'belt' found?: false
 8Word 'better' found?: true
 9Word 'bet' found?: false
10Word 'betting' found?: true
11Word 'car' found?: true
12Word 'cart' found?: true
13Word 'call' found?: false
14Word 'called' found?: true
How the above code works:

Deletion can be a bit hard to understand so we’ll walkthrough each step for each word we are going to delete.

Here’s the initial state of the trie.

t-all-rk-blt-eerr-otto-ting-rt--called--

Deleting “bet”

  • We start with the root node and depth 0 and recursively pass the word down the trie.
  • Root node is not nil and the depth [0] != len(‘bet’) [3]
  • We extract the child node at depth = 0 i.e. ‘b’, increment the depth by 1 and call the deleteHelper function for node ‘b’.
  • Node ‘b’ is not nil and the depth [1] != len(‘bet’) [3]
  • We extract the child node at depth = 1 i.e. ’e’, increment the depth by 1 and call the deleteHelper function for node ’e’.
  • Node ’e’ is not nil and the depth[2] != len(‘bet’) [3]
  • We extract the child node at depth = 2 i.e. ’t’, increment the depth by 1 and call the deleteHelper function for node ’t’.
  • Node ’t’ is not nil and the depth[3] == len(‘bet’) [3] which means that we have found all the characters matching the word in the trie.
  • Node ’t’ has endOfWord flag set to true which means the word “bet” exists in the trie.
  • We set the endOfWord flag to false.
  • Node ’t’ has children which means there are other words that start with prefix “bet”. So we return false, so that the calling function doesn’t delete node ’t’.
  • False is returned recursively so each calling chain doesn’t delete the node.

State of Trie after “bet” has been deleted.

t-all-rk-blt-eerr-ottoting-rt--called--

Deleting “belt”

  • We start with the root node and depth 0 and recursively pass the word down the trie.
  • Root node is not nil and the depth [0] != len(‘belt’) [4]
  • We extract the child node at depth = 0 i.e. ‘b’, increment the depth by 1 and call the deleteHelper function for node ‘b’.
  • Node ‘b’ is not nil and the depth [1] != len(‘belt’) [4]
  • We extract the child node at depth = 1 i.e. ’e’, increment the depth by 1 and call the deleteHelper function for node ’e’.
  • Node ’e’ is not nil and the depth[2] != len(‘belt’) [4]
  • We extract the child node at depth = 2 i.e. ’l’, increment the depth by 1 and call the deleteHelper function for node ’l’.
  • Node ’l’ is not nil and the depth[3] != len(‘belt’) [4]
  • We extract the child node at depth = 3 i.e. ’t’, increment the depth by 1 and call the deleteHelper function for node ’t’.
  • Node ’t’ is not nil and the depth[4] == len(‘belt’) [4] which means that we have found all the characters matching the word in the trie.
  • Node ’t’ has endOfWord flag set to true which means the word “belt” exists in the trie.
  • We set the endOfWord flag to false.
t-all-rk-blteerr-ottoting-rt--called--
  • Node ’t’ has no children which means there are no other words that start with prefix “belt”. So we return true.
  • The calling function then deletes node ’t’
t-all-rk-bleerr-ottoting-rt--called--
  • Current node ’l’ doesn’t have endOfWord flag set to true which means there is no word “bel” stored in the trie. Also it is a leaf node after deletion of it’s only child ’t’. True is returned to the calling function.
  • The calling function then deletes node ’l’
t-all-brk-er-etting-rootrt--called--
  • Current node ’e’ doesn’t have endOfWord flag set to true which means there is no word “be” stored in the trie. Also it is not a leaf node since the words “better” and “betting” are still present in the trie. False is returned to the calling function.
  • The calling function then doesn’t delete ’e’ node and also recursively return false to the calling function till root and no other node is delete on the path.

Deleting “call”

  • Similar to deleting “bet”. The endOfWord flag for the 2nd l is set to false and no nodes on the path deleted because the word “called” is still in the trie.
t-all-brk-er-etting-rootrt--called-

Conclusion

In this post, we’ve explored the Trie data structure, a powerful tool for efficiently storing and searching words, especially when dealing with common prefixes. We walked through the basics of Tries and implemented essential operations in Golang, including insertion, search, prefix search, and deletion. By adding functionalities like returning words that match a given prefix, we showcased how Tries can be extended to support features like autocomplete.

Whether you’re building a search engine, implementing autocomplete, or just optimizing text storage, understanding and using Tries can significantly enhance the performance and capabilities of your applications.