mirror of
https://github.com/mii443/AzooKeyKanaKanjiConverter.git
synced 2025-08-22 15:05:26 +00:00
[Experimental] Zenzai (#92)
* experimental rinna integration * Update impl * update * Bump swift-actions/setup-swift from 1 to 2 Bumps [swift-actions/setup-swift](https://github.com/swift-actions/setup-swift) from 1 to 2. - [Release notes](https://github.com/swift-actions/setup-swift/releases) - [Commits](https://github.com/swift-actions/setup-swift/compare/v1...v2) --- updated-dependencies: - dependency-name: swift-actions/setup-swift dependency-type: direct:production update-type: version-update:semver-major ... Signed-off-by: dependabot[bot] <support@github.com> * change test * change impl * take gpt2 weight as option * don't use async * support status check * enhance error * avoid percent encode * update * GPT-2 based kana-kanji conversion is now perfectly workinggit statusgit status * fix a bug * Rename gpt2/llama -> zenz * cleanup * internal apiを綺麗にした * cleanup experimental commands * update * partially support incremental input using cache * fix names * fix bug * support roman2kana * cleanup * fix minor bugs * improve logic * fix minor bug * fix minor bug * fix minor bug * optimize * optimize performance * Optimize cache hit * cli: add anco session command * fix cache hit bugs * improve session commands * maybe this will work better for incremental input environment * speed up zenzai by using n_best alternatives * update zenz context * adding no_typo api * add inference limit * fix bug * reset install_cli * make package buildable -- but llama.cpp features just do not work at this point because metal is not preprocessed * add proper availability checks * change macOS minimum version * fix several problems * code cleanup * enable ubuntu build * fix build error * fix ubuntu build * fix borrowing * update install_cli.sh --------- Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
This commit is contained in:
Miwa
GitHub
@ -10,7 +10,7 @@
|
||||
import XCTest
|
||||
|
||||
final class LOUDSTests: XCTestCase {
|
||||
static var resourceURL = Bundle.module.resourceURL!.standardizedFileURL.appendingPathComponent("DictionaryMock", isDirectory: true)
|
||||
static let resourceURL = Bundle.module.resourceURL!.standardizedFileURL.appendingPathComponent("DictionaryMock", isDirectory: true)
|
||||
func requestOptions() -> ConvertRequestOptions {
|
||||
var options: ConvertRequestOptions = .default
|
||||
options.dictionaryResourceURL = Self.resourceURL
|
||||
|
||||
@ -19,8 +19,8 @@ final class ConverterTests: XCTestCase {
|
||||
|
||||
func requestOptions() -> ConvertRequestOptions {
|
||||
.withDefaultDictionary(
|
||||
N_best: 5,
|
||||
requireJapanesePrediction: true,
|
||||
N_best: 10,
|
||||
requireJapanesePrediction: false,
|
||||
requireEnglishPrediction: false,
|
||||
keyboardLanguage: .ja_JP,
|
||||
typographyLetterCandidate: false,
|
||||
@ -38,46 +38,41 @@ final class ConverterTests: XCTestCase {
|
||||
}
|
||||
|
||||
func testFullConversion() async throws {
|
||||
await MainActor.run {
|
||||
do {
|
||||
let converter = KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
c.insertAtCursorPosition("あずーきーはしんじだいのきーぼーどあぷりです", inputStyle: .direct)
|
||||
let results = converter.requestCandidates(c, options: requestOptions())
|
||||
XCTAssertEqual(results.mainResults.first?.text, "azooKeyは新時代のキーボードアプリです")
|
||||
}
|
||||
do {
|
||||
let converter = KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
c.insertAtCursorPosition("ようしょうきからてにすすいえいやきゅうしょうりんじけんぽうなどさまざまなすぽーつをけいけんしながらそだちしょうがっこうじだいはろさんぜるすきんこうにたいざいしておりごるふやてにすをならっていた", inputStyle: .direct)
|
||||
let results = converter.requestCandidates(c, options: requestOptions())
|
||||
do {
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
c.insertAtCursorPosition("あずーきーはしんじだいのきーぼーどあぷりです", inputStyle: .direct)
|
||||
let results = await converter.requestCandidates(c, options: requestOptions())
|
||||
XCTAssertEqual(results.mainResults.first?.text, "azooKeyは新時代のキーボードアプリです")
|
||||
}
|
||||
do {
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
c.insertAtCursorPosition("ようしょうきからてにすすいえいやきゅうしょうりんじけんぽうなどさまざまなすぽーつをけいけんしながらそだちしょうがっこうじだいはろさんぜるすきんこうにたいざいしておりごるふやてにすをならっていた", inputStyle: .direct)
|
||||
let results = await converter.requestCandidates(c, options: requestOptions())
|
||||
XCTAssertEqual(results.mainResults.first?.text, "幼少期からテニス水泳野球少林寺拳法など様々なスポーツを経験しながら育ち小学校時代はロサンゼルス近郊に滞在しておりゴルフやテニスを習っていた")
|
||||
}
|
||||
}
|
||||
|
||||
// 1文字ずつ変換する
|
||||
// memo: 内部実装としては別のモジュールが呼ばれるのだが、それをテストする方法があまりないかもしれない
|
||||
func testGradualConversion() async throws {
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
let text = "ようしょうきからてにすすいえいやきゅうしょうりんじけんぽうなどさまざまなすぽーつをけいけんしながらそだちしょうがっこうじだいはろさんぜるすきんこうにたいざいしておりごるふやてにすをならっていた"
|
||||
for char in text {
|
||||
c.insertAtCursorPosition(String(char), inputStyle: .direct)
|
||||
let results = await converter.requestCandidates(c, options: requestOptions())
|
||||
if c.input.count == text.count {
|
||||
XCTAssertEqual(results.mainResults.first?.text, "幼少期からテニス水泳野球少林寺拳法など様々なスポーツを経験しながら育ち小学校時代はロサンゼルス近郊に滞在しておりゴルフやテニスを習っていた")
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 1文字ずつ変換する
|
||||
// memo: 内部実装としては別のモジュールが呼ばれるのだが、それをテストする方法があまりないかもしれない
|
||||
func testGradualConversion() async throws {
|
||||
await MainActor.run {
|
||||
let converter = KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
let text = "ようしょうきからてにすすいえいやきゅうしょうりんじけんぽうなどさまざまなすぽーつをけいけんしながらそだちしょうがっこうじだいはろさんぜるすきんこうにたいざいしておりごるふやてにすをならっていた"
|
||||
for char in text {
|
||||
c.insertAtCursorPosition(String(char), inputStyle: .direct)
|
||||
let results = converter.requestCandidates(c, options: requestOptions())
|
||||
if c.input.count == text.count {
|
||||
XCTAssertEqual(results.mainResults.first?.text, "幼少期からテニス水泳野球少林寺拳法など様々なスポーツを経験しながら育ち小学校時代はロサンゼルス近郊に滞在しておりゴルフやテニスを習っていた")
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 1文字ずつ変換する
|
||||
// memo: 内部実装としては別のモジュールが呼ばれるのだが、それをテストする方法があまりないかもしれない
|
||||
func testRoman2KanaGradualConversion() async throws {
|
||||
await MainActor.run {
|
||||
let converter = KanaKanjiConverter()
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
let text = "youshoukikaratenisusuieiyakyuushourinjikenpounadosamazamanasupoーtuwokeikennsinagarasodatishougakkouzidaiharosanzerusukinkounitaizaisiteorigoruhuyatenisuwonaratteita"
|
||||
// 許容される変換結果
|
||||
@ -87,19 +82,17 @@ final class ConverterTests: XCTestCase {
|
||||
]
|
||||
for char in text {
|
||||
c.insertAtCursorPosition(String(char), inputStyle: .roman2kana)
|
||||
let results = converter.requestCandidates(c, options: requestOptions())
|
||||
let results = await converter.requestCandidates(c, options: requestOptions())
|
||||
if c.input.count == text.count {
|
||||
XCTAssertTrue(possibles.contains(results.mainResults.first!.text))
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 2,3文字ずつ変換する
|
||||
// memo: 内部実装としては別のモジュールが呼ばれるのだが、それをテストする方法があまりないかもしれない
|
||||
func testSemiGradualConversion() async throws {
|
||||
await MainActor.run {
|
||||
let converter = KanaKanjiConverter()
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
let text = "ようしょうきからてにすすいえいやきゅうしょうりんじけんぽうなどさまざまなすぽーつをけいけんしながらそだちしょうがっこうじだいはろさんぜるすきんこうにたいざいしておりごるふやてにすをならっていた"
|
||||
var leftIndex = text.startIndex
|
||||
@ -110,44 +103,40 @@ final class ConverterTests: XCTestCase {
|
||||
let rightIndex = text.index(leftIndex, offsetBy: count, limitedBy: text.endIndex) ?? text.endIndex
|
||||
let prefix = String(text[leftIndex ..< rightIndex])
|
||||
c.insertAtCursorPosition(prefix, inputStyle: .direct)
|
||||
let results = converter.requestCandidates(c, options: requestOptions())
|
||||
let results = await converter.requestCandidates(c, options: requestOptions())
|
||||
leftIndex = rightIndex
|
||||
if rightIndex == text.endIndex {
|
||||
XCTAssertEqual(results.mainResults.first?.text, "幼少期からテニス水泳野球少林寺拳法など様々なスポーツを経験しながら育ち小学校時代はロサンゼルス近郊に滞在しておりゴルフやテニスを習っていた")
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 1文字ずつ入力するが、時折削除を行う
|
||||
// memo: 内部実装としてはdeleted_last_nのテストを意図している
|
||||
func testGradualConversionWithDelete() async throws {
|
||||
await MainActor.run {
|
||||
let converter = KanaKanjiConverter()
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
let text = Array("ようしょうきからてにすすいえいやきゅうしょうりんじけんぽうなどさまざまなすぽーつをけいけんしながらそだちしょうがっこうじだいはろさんぜるすきんこうにたいざいしておりごるふやてにすをならっていた")
|
||||
let deleteIndices = [1, 4, 8, 10, 15, 18, 20, 21, 23, 25, 26, 28, 29, 33, 34, 37, 39, 40, 42, 44, 45, 49, 51, 54, 58, 60, 62, 64, 67, 69, 70, 75, 80]
|
||||
for (i, char) in text.enumerated() {
|
||||
c.insertAtCursorPosition(String(char), inputStyle: .direct)
|
||||
let results = converter.requestCandidates(c, options: requestOptions())
|
||||
let results = await converter.requestCandidates(c, options: requestOptions())
|
||||
if deleteIndices.contains(i) {
|
||||
let count = i % 3 + 1
|
||||
c.deleteBackwardFromCursorPosition(count: count)
|
||||
_ = converter.requestCandidates(c, options: requestOptions())
|
||||
_ = await converter.requestCandidates(c, options: requestOptions())
|
||||
|
||||
c.insertAtCursorPosition(String(text[i - count + 1 ... i]), inputStyle: .direct)
|
||||
_ = converter.requestCandidates(c, options: requestOptions())
|
||||
_ = await converter.requestCandidates(c, options: requestOptions())
|
||||
}
|
||||
if c.input.count == text.count {
|
||||
XCTAssertEqual(results.mainResults.first?.text, "幼少期からテニス水泳野球少林寺拳法など様々なスポーツを経験しながら育ち小学校時代はロサンゼルス近郊に滞在しておりゴルフやテニスを習っていた")
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 必ず正解すべきテストケース
|
||||
func testMustCases() async throws {
|
||||
await MainActor.run {
|
||||
// ダイレクト入力
|
||||
do {
|
||||
let cases: [(input: String, expect: String)] = [
|
||||
@ -162,19 +151,19 @@ final class ConverterTests: XCTestCase {
|
||||
var options = requestOptions()
|
||||
options.requireJapanesePrediction = false
|
||||
for (input, expect) in cases {
|
||||
let converter = KanaKanjiConverter()
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
sequentialInput(&c, sequence: input, inputStyle: .direct)
|
||||
let results = converter.requestCandidates(c, options: options)
|
||||
let results = await converter.requestCandidates(c, options: options)
|
||||
XCTAssertEqual(results.mainResults.first?.text, expect)
|
||||
}
|
||||
// gradual input
|
||||
for (input, expect) in cases {
|
||||
let converter = KanaKanjiConverter()
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
for char in input {
|
||||
c.insertAtCursorPosition(String(char), inputStyle: .direct)
|
||||
let results = converter.requestCandidates(c, options: options)
|
||||
let results = await converter.requestCandidates(c, options: options)
|
||||
if c.input.count == input.count {
|
||||
XCTAssertEqual(results.mainResults.first?.text, expect)
|
||||
}
|
||||
@ -193,33 +182,31 @@ final class ConverterTests: XCTestCase {
|
||||
var options = requestOptions()
|
||||
options.requireJapanesePrediction = false
|
||||
for (input, expect) in cases {
|
||||
let converter = KanaKanjiConverter()
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
sequentialInput(&c, sequence: input, inputStyle: .roman2kana)
|
||||
let results = converter.requestCandidates(c, options: options)
|
||||
let results = await converter.requestCandidates(c, options: options)
|
||||
XCTAssertEqual(results.mainResults.first?.text, expect)
|
||||
}
|
||||
|
||||
// gradual input
|
||||
for (input, expect) in cases {
|
||||
let converter = KanaKanjiConverter()
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
for char in input {
|
||||
c.insertAtCursorPosition(String(char), inputStyle: .roman2kana)
|
||||
let results = converter.requestCandidates(c, options: options)
|
||||
let results = await converter.requestCandidates(c, options: options)
|
||||
if c.input.count == input.count {
|
||||
XCTAssertEqual(results.mainResults.first?.text, expect)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 変換結果が比較的一意なテストケースを無数に持ち、一定の割合を正解することを要求する
|
||||
// 辞書を更新した結果性能が悪化したら気付ける
|
||||
func testAccuracy() async throws {
|
||||
await MainActor.run {
|
||||
let cases: [(input: String, expect: [String])] = [
|
||||
("3がつ8にち", ["3月8日"]),
|
||||
("いっていのわりあい", ["一定の割合"]),
|
||||
@ -275,10 +262,10 @@ final class ConverterTests: XCTestCase {
|
||||
|
||||
var score: Double = 0
|
||||
for (input, expect) in cases {
|
||||
let converter = KanaKanjiConverter()
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
c.insertAtCursorPosition(input, inputStyle: .direct)
|
||||
let results = converter.requestCandidates(c, options: requestOptions())
|
||||
let results = await converter.requestCandidates(c, options: requestOptions())
|
||||
|
||||
if expect.contains(results.mainResults[0].text) {
|
||||
score += 1
|
||||
@ -291,14 +278,12 @@ final class ConverterTests: XCTestCase {
|
||||
let accuracy = score / Double(cases.count)
|
||||
print("\(#function) Result: accuracy \(accuracy), score \(score), count \(cases.count)")
|
||||
XCTAssertGreaterThan(accuracy, 0.7) // 0.7 < acuracy
|
||||
}
|
||||
}
|
||||
|
||||
// 変換結果が比較的一意なテストケースを無数に持ち、一定の割合を正解することを要求する
|
||||
// 辞書を更新した結果性能が悪化したら気付ける
|
||||
// 口語表現を中心にテストする
|
||||
func testVerbalAccuracy() async throws {
|
||||
await MainActor.run {
|
||||
let cases: [(input: String, expect: [String])] = [
|
||||
("うわああああ、まじか", ["うわああああ、マジか", "うわああああ、まじか"]),
|
||||
("は？", ["は？"]),
|
||||
@ -326,10 +311,10 @@ final class ConverterTests: XCTestCase {
|
||||
|
||||
var score: Double = 0
|
||||
for (input, expect) in cases {
|
||||
let converter = KanaKanjiConverter()
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
c.insertAtCursorPosition(input, inputStyle: .direct)
|
||||
let results = converter.requestCandidates(c, options: requestOptions())
|
||||
let results = await converter.requestCandidates(c, options: requestOptions())
|
||||
|
||||
if expect.contains(results.mainResults[0].text) {
|
||||
score += 1
|
||||
@ -342,12 +327,10 @@ final class ConverterTests: XCTestCase {
|
||||
let accuracy = score / Double(cases.count)
|
||||
print("\(#function) Result: accuracy \(accuracy), score \(score), count \(cases.count)")
|
||||
XCTAssertGreaterThan(accuracy, 0.7) // 0.7 < acuracy
|
||||
}
|
||||
}
|
||||
|
||||
/// MIDベースの文節単位計算でどれだけ同音異義語の判断が向上しているか確認する。
|
||||
func testMeaningBasedConversionAccuracy() async throws {
|
||||
await MainActor.run {
|
||||
let cases: [(input: String, expect: String)] = [
|
||||
("しょうぼう、しょうか、ほのお", "消防、消火、炎"),
|
||||
("いえき、しょうか、こうそ", "胃液、消化、酵素"),
|
||||
@ -627,12 +610,12 @@ final class ConverterTests: XCTestCase {
|
||||
|
||||
var score: Double = 0
|
||||
for (input, expect) in cases {
|
||||
let converter = KanaKanjiConverter()
|
||||
let converter = await KanaKanjiConverter()
|
||||
var c = ComposingText()
|
||||
c.insertAtCursorPosition(input, inputStyle: .direct)
|
||||
var options = requestOptions()
|
||||
options.requireJapanesePrediction = false
|
||||
let results = converter.requestCandidates(c, options: options)
|
||||
let results = await converter.requestCandidates(c, options: options)
|
||||
|
||||
if results.mainResults[0].text == expect {
|
||||
score += 1
|
||||
@ -645,10 +628,9 @@ final class ConverterTests: XCTestCase {
|
||||
let accuracy = score / Double(cases.count)
|
||||
print("\(#function) Result: accuracy \(accuracy), score \(score), count \(cases.count)")
|
||||
XCTAssertGreaterThan(accuracy, 0.7) // 0.7 < accuracy
|
||||
}
|
||||
}
|
||||
|
||||
#if os(macOS) || os(iOS) || os(watchOS) || os(tvOS) || os(visionOS)
|
||||
#if os(macOS) || os(iOS) || os(watchOS) || os(tvOS) || os(visionOS)
|
||||
func testMozcEvaluationData() async throws {
|
||||
// ダウンロードするURL
|
||||
let urlString = "https://raw.githubusercontent.com/google/mozc/master/src/data/dictionary_oss/evaluation.tsv"
|
||||
@ -729,7 +711,7 @@ final class ConverterTests: XCTestCase {
|
||||
XCTAssertTrue(mozcScore < azooKeyScore)
|
||||
}
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
enum MozcCommand: Equatable {
|
||||
/// 変換に`arg`が現れる
|
||||
|
||||
Reference in New Issue
Block a user