WIP: implementing zero query prediction

Sources/KanaKanjiConverterModule/Converter/KanaKanjiConverter.swift

@@ -414,15 +414,12 @@ import SwiftUtils
 
         // 文全体変換5件と予測変換3件を混ぜてベスト8を出す
         let best8 = getUniqueCandidate(sentence_candidates.chained(prediction_candidates)).sorted {$0.value > $1.value}
-        // ゼロヒント予測変換
-        let zeroHintPrediction_candidates = converter.getZeroHintPredictionCandidates(preparts: best8, N_best: 3)
         // その他のトップレベル変換（先頭に表示されうる変換候補）
         let toplevel_additional_candidate = self.getTopLevelAdditionalCandidate(inputData, options: options)
         // best8、foreign_candidates、zeroHintPrediction_candidates、toplevel_additional_candidateを混ぜて上位5件を取得する
         let full_candidate = getUniqueCandidate(
             best8
                 .chained(foreign_candidates)
-                .chained(zeroHintPrediction_candidates)
                 .chained(toplevel_additional_candidate)
         ).min(count: 5, sortedBy: {$0.value > $1.value})
         // 重複のない変換候補を作成するための集合
@@ -570,6 +567,11 @@ import SwiftUtils
 
     }
 
+    /// 2つの連続する`Candidate`をマージする
+    public func mergeCandidates(_ left: Candidate, _ right: Candidate) -> Candidate {
+        converter.mergeCandidates(left, right)
+    }
+
     /// 外部から呼ばれる変換候補を要求する関数。
     /// - Parameters:
     ///   - inputData: 変換対象のInputData。
@@ -591,4 +593,17 @@ import SwiftUtils
 
         return self.processResult(inputData: inputData, result: result, options: options)
     }
+    
+    /// 変換確定後の予測変換候補を要求する関数
+    public func requestPredictionCandidates(leftSideCandidate: Candidate, options: ConvertRequestOptions) -> [Candidate] {
+        // ゼロヒント予測変換に基づく候補を列挙
+        let zeroHintResults = self.converter.getZeroHintPredictionCandidates(preparts: [leftSideCandidate], N_best: 10)
+        // 予測変換に基づく候補を列挙
+        // TODO: implement
+        // 学習・ユーザ辞書に基づく候補を列挙
+        // TODO: implement
+        // 絵文字、記号類を列挙
+        // TODO: implement
+        return zeroHintResults
+    }
 }

Sources/KanaKanjiConverterModule/DicdataStore/DicdataStore.swift

@@ -29,7 +29,6 @@ public final class DicdataStore {
     private var importedLoudses: Set<String> = []
     private var charsID: [Character: UInt8] = [:]
     private var learningManager = LearningManager()
-    private var zeroHintPredictionDicdata: [DicdataElement]?
 
     private var osUserDict: [DicdataElement] = []
 
@@ -361,20 +360,15 @@ public final class DicdataStore {
         }
     }
 
-    func getZeroHintPredictionDicdata() -> [DicdataElement] {
-        if let dicdata = self.zeroHintPredictionDicdata {
-            return dicdata
-        }
+    func getZeroHintPredictionDicdata(lastRcid: Int) -> [DicdataElement] {
         do {
-            let csvString = try String(contentsOf: requestOptions.dictionaryResourceURL.appendingPathComponent("p/p_null.csv", isDirectory: false), encoding: String.Encoding.utf8)
+            let csvString = try String(contentsOf: requestOptions.dictionaryResourceURL.appendingPathComponent("p/pc_\(lastRcid).csv", isDirectory: false), encoding: .utf8)
             let csvLines = csvString.split(separator: "\n")
             let csvData = csvLines.map {$0.split(separator: ",", omittingEmptySubsequences: false)}
             let dicdata: [DicdataElement] = csvData.map {self.parseLoudstxt2FormattedEntry(from: $0)}
-            self.zeroHintPredictionDicdata = dicdata
             return dicdata
         } catch {
             debug(error)
-            self.zeroHintPredictionDicdata = []
             return []
         }
     }

Sources/KanaKanjiConverterModule/Kana2Kanji/Kana2Kanji.swift

@@ -44,6 +44,29 @@ struct Kana2Kanji {
         )
     }
 
+    public func mergeCandidates(_ left: Candidate, _ right: Candidate) -> Candidate {
+        guard let leftLast = left.data.last, let rightFirst = right.data.first else {
+            return Candidate(
+                text: left.text + right.text,
+                value: left.value + right.value,
+                correspondingCount: left.correspondingCount + right.correspondingCount,
+                lastMid: right.lastMid,
+                data: left.data + right.data
+            )
+        }
+        let ccValue = self.dicdataStore.getCCValue(leftLast.lcid, rightFirst.lcid)
+        let includeMMValueCalculation = DicdataStore.includeMMValueCalculation(rightFirst)
+        let mmValue = includeMMValueCalculation ? self.dicdataStore.getMMValue(left.lastMid, rightFirst.mid):.zero
+        let newValue = left.value + mmValue + ccValue + right.value
+        return Candidate(
+            text: left.text + right.text,
+            value: newValue,
+            correspondingCount: left.correspondingCount + right.correspondingCount,
+            lastMid: right.lastMid,
+            data: left.data + right.data
+        )
+    }
+
     /// 入力がない状態から、妥当な候補を探す
     /// - parameters:
     ///   - preparts: Candidate列。以前確定した候補など
@@ -53,54 +76,37 @@ struct Kana2Kanji {
     /// - note:
     ///   「食べちゃ-てる」「食べちゃ-いる」などの間抜けな候補を返すことが多いため、学習によるもの以外を無効化している。
     func getZeroHintPredictionCandidates(preparts: some Collection<Candidate>, N_best: Int) -> [Candidate] {
-        // let dicdata = self.dicdataStore.getZeroHintPredictionDicdata()
         var result: [Candidate] = []
-        /*
-         result.reserveCapacity(N_best + 1)
-         preparts.forEach{candidate in
-         dicdata.forEach{data in
-         let ccValue = self.dicdataStore.getCCValue(candidate.rcid, data.lcid)
-         let isInposition = DicdataStore.isInposition(data)
-         let mmValue = isInposition ? self.dicdataStore.getMMValue(candidate.lastMid, data.mid):0.0
-         let wValue = data.value()
-         let newValue = candidate.value + mmValue + ccValue + wValue
-         //追加すべきindexを取得する
-         let lastindex = (result.lastIndex(where: {$0.value >= newValue}) ?? -1) + 1
-         if lastindex >= N_best{
-         return
-         }
-         var nodedata = candidate.data
-         nodedata.append(data)
-
-         let candidate = Candidate(text: candidate.text + data.string, value: newValue, correspondingCount: candidate.correspondingCount, rcid: data.rcid, lastMid: isInposition ? data.mid:candidate.lastMid, data: nodedata)
-         result.insert(candidate, at: lastindex)
-         //カウントがオーバーしている場合は除去する
-         if result.count == N_best &+ 1{
-         result.removeLast()
-         }
-         }
-         }
-         */
         for candidate in preparts {
             if let last = candidate.data.last {
-                let nexts = [(DicdataElement, Int)]()
-                for (data, count) in nexts where count > 1 {
+                let dicdata = self.dicdataStore.getZeroHintPredictionDicdata(lastRcid: last.rcid)
+                for data in dicdata {
                     let ccValue = self.dicdataStore.getCCValue(last.rcid, data.lcid)
                     let includeMMValueCalculation = DicdataStore.includeMMValueCalculation(data)
                     let mmValue = includeMMValueCalculation ? self.dicdataStore.getMMValue(candidate.lastMid, data.mid):.zero
                     let wValue = data.value()
-                    let bonus = PValue(count * 1)
-                    let newValue = candidate.value + mmValue + ccValue + wValue + bonus
+                    let newValue = candidate.value + mmValue + ccValue + wValue
+
+                    // 追加すべきindexを取得する
+                    let lastindex: Int = (result.lastIndex(where: {$0.value >= newValue}) ?? -1) + 1
+                    if lastindex == N_best {
+                        continue
+                    }
+                    // データを作成する
                     var nodedata = candidate.data
                     nodedata.append(data)
                     let candidate = Candidate(
-                        text: candidate.text + data.word,
-                        value: newValue,
-                        correspondingCount: candidate.correspondingCount,
-                        lastMid: includeMMValueCalculation ? data.mid:candidate.lastMid,
-                        data: nodedata
+                        text: data.word,
+                        value: data.value(),
+                        correspondingCount: data.ruby.count,
+                        lastMid: data.mid,
+                        data: [data]
                     )
-                    result.append(candidate)
+                    // カウントがオーバーしている場合は除去する
+                    if result.count >= N_best {
+                        result.removeLast()
+                    }
+                    result.insert(candidate, at: lastindex)
                 }
             }
         }