-
Notifications
You must be signed in to change notification settings - Fork 46
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Merge pull request #12 from jordanekay/naive-bayes
Convert naive Bayes classifier to Swift
- Loading branch information
Showing
9 changed files
with
225 additions
and
293 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,16 @@ | ||
| // | ||
| // Functions.swift | ||
| // Parsimmon | ||
| // | ||
| // Created by Jordan Kay on 2/18/15. | ||
| // | ||
| // | ||
|
|
||
| import Foundation | ||
|
|
||
| func argmax<T, U: Comparable>(elements: [(T, U)]) -> T? { | ||
| if let start = elements.first { | ||
| return elements.reduce(start) { $0.1 > $1.1 ? $0 : $1 }.0 | ||
| } | ||
| return nil | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,158 @@ | ||
| // | ||
| // NaiveBayesClassifier.swift | ||
| // Parsimmon | ||
| // | ||
| // Created by Jordan Kay on 2/18/15. | ||
| // | ||
| // | ||
|
|
||
| /** | ||
| ## Sample usage | ||
| let NaiveBayesClassifier classifier = NaiveBayesClassifier() | ||
| Train the classifier with some ham examples. | ||
| classifier.trainWithText("nom nom ham", category: "ham") | ||
| classifier.trainWithText("make sure to get the ham", category: "ham") | ||
| classifier.trainWithText("please put the eggs in the fridge", category: "ham") | ||
| Train the classifier with some spam examples. | ||
| classifier.trainWithText("spammy spam spam", category: "spam") | ||
| classifier.trainWithText("what does the fox say?", category: "spam") | ||
| classifier.trainWithText("and fish go blub", category: "spam") | ||
| Classify some new text. Is it ham or spam? In practice, you'd want to train with more examples first. | ||
| let firstExample = "use the eggs in the fridge." | ||
| let secondExample = "what does the fish say?" | ||
| println("'\(firstExample)' => \(classifier.classify(firstExample)") | ||
| println("'\(secondExample)' => \(classifier.classify(secondExample)") | ||
| Output: | ||
| 'use the eggs in the fridge.' => ham | ||
| 'what does the fish say?' => spam | ||
| */ | ||
|
|
||
| import Foundation | ||
|
|
||
| private let smoothingParameter = 1.0 | ||
|
|
||
| public class NaiveBayesClassifier: NSObject { | ||
| public typealias Word = String | ||
| public typealias Category = String | ||
|
|
||
| private let tokenizer: ParsimmonTokenizer | ||
|
|
||
| private var categoryOccurrences: [Category: Int] = [:] | ||
| private var wordOccurrences: [Word: [Category: Int]] = [:] | ||
| private var trainingCount = 0 | ||
| private var wordCount = 0 | ||
|
|
||
| public init(tokenizer: ParsimmonTokenizer) { | ||
| self.tokenizer = tokenizer | ||
| } | ||
|
|
||
| public convenience override init() { | ||
| self.init(tokenizer: ParsimmonTokenizer()) | ||
| } | ||
|
|
||
| // MARK: - Training | ||
|
|
||
| /** | ||
| Trains the classifier with text and its category. | ||
| @param text The text | ||
| @param category The category of the text | ||
| */ | ||
| public func trainWithText(text: String, category: Category) { | ||
| let tokens = tokenizer.tokenize(text) | ||
| trainWithTokens(tokens, category: category) | ||
| } | ||
|
|
||
| /** | ||
| Trains the classifier with tokenized text and its category. | ||
| This is useful if you wish to use your own tokenization method. | ||
| @param tokens The tokenized text | ||
| @param category The category of the text | ||
| */ | ||
| public func trainWithTokens(tokens: [Word], category: Category) { | ||
| let words = Set(tokens) | ||
| for word in words { | ||
| incrementWord(word, category: category) | ||
| } | ||
| incrementCategory(category) | ||
| trainingCount++ | ||
| } | ||
|
|
||
| // MARK: - Classifying | ||
|
|
||
| /** | ||
| Classifies the given text based on its training data. | ||
| @param text The text to classify | ||
| @return The category classification | ||
| */ | ||
| public func classify(text: String) -> Category? { | ||
| let tokens = tokenizer.tokenize(text) | ||
| return classifyTokens(tokens) | ||
| } | ||
|
|
||
| /** | ||
| Classifies the given tokenized text based on its training data. | ||
| @param text The tokenized text to classify | ||
| @return The category classification if one was found, or nil if one wasn’t | ||
| */ | ||
| public func classifyTokens(tokens: [Word]) -> Category? { | ||
| // Compute argmax_cat [log(P(C=cat)) + sum_token(log(P(W=token|C=cat)))] | ||
| return argmax(map(categoryOccurrences) { (category, count) -> (Category, Double) in | ||
| let pCategory = P(category) | ||
| let score = tokens.reduce(log(pCategory)) { [wordCount] (total, token) in | ||
| total + log((self.P(category, token) + smoothingParameter) / (pCategory + smoothingParameter + Double(wordCount))) | ||
| } | ||
| return (category, score) | ||
| }) | ||
| } | ||
|
|
||
| // MARK: - Probabilites | ||
|
|
||
| private func P(category: Category, _ word: Word) -> Double { | ||
| if let occurrences = wordOccurrences[word] { | ||
| let count = occurrences[category] ?? 0 | ||
| return Double(count) / Double(trainingCount) | ||
| } | ||
| return 0.0 | ||
| } | ||
|
|
||
| private func P(category: Category) -> Double { | ||
| return Double(totalOccurrencesOfCategory(category)) / Double(trainingCount) | ||
| } | ||
|
|
||
| // MARK: - Counting | ||
|
|
||
| private func incrementWord(word: Word, category: Category) { | ||
| if wordOccurrences[word] == nil { | ||
| wordCount++ | ||
| wordOccurrences[word] = [:] | ||
| } | ||
|
|
||
| let count = wordOccurrences[word]?[category] ?? 0 | ||
| wordOccurrences[word]?[category] = count + 1 | ||
| } | ||
|
|
||
| private func incrementCategory(category: Category) { | ||
| categoryOccurrences[category] = totalOccurrencesOfCategory(category) + 1 | ||
| } | ||
|
|
||
| private func totalOccurrencesOfWord(word: Word) -> Int { | ||
| if let occurrences = wordOccurrences[word] { | ||
| return Array(occurrences.values).reduce(0, combine: +) | ||
| } | ||
| return 0 | ||
| } | ||
|
|
||
| private func totalOccurrencesOfCategory(category: Category) -> Int { | ||
| return categoryOccurrences[category] ?? 0 | ||
| } | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -22,4 +22,3 @@ | |
|
|
||
| #import "ParsimmonTagger.h" | ||
| #import "ParsimmonLemmatizer.h" | ||
| #import "ParsimmonNaiveBayesClassifier.h" | ||
This file was deleted.
Oops, something went wrong.
Oops, something went wrong.