Add executable 50-point household checklist #365
@@ -250,6 +250,9 @@ func semanticScores(tokens []string, hasQuestionMark bool, explicitCorrection bo
|
||||
if explicitCorrection {
|
||||
scores[interaction.SpeechActCorrection] += 6
|
||||
}
|
||||
if containsStemInTokens(tokens, "опечат", "ошибк", "исправлен") && containsStemInTokens(tokens, "исправ", "поправ") {
|
||||
scores[interaction.SpeechActCorrection] += 6
|
||||
}
|
||||
first := ""
|
||||
if len(tokens) > 0 {
|
||||
first = tokens[0]
|
||||
@@ -496,19 +499,35 @@ func capabilityHintsFor(primary, secondary interaction.SpeechAct, tokens []strin
|
||||
add("WEB_RESEARCH")
|
||||
}
|
||||
if isCreativeRequestTokens(tokens) {
|
||||
if isImageCreationRequestTokens(tokens) {
|
||||
// A correction that quotes a creative phrase is still a correction;
|
||||
// do not execute the quoted phrase as a side effect.
|
||||
if isImageCreationRequestTokens(tokens) && primary != interaction.SpeechActCorrection && secondary != interaction.SpeechActCorrection {
|
||||
add("IMAGE_GENERATION")
|
||||
} else {
|
||||
add("COGNITIVE_REASONING")
|
||||
}
|
||||
}
|
||||
if containsStemInTokens(tokens, "голосов", "аудио", "послуш") {
|
||||
add("SPEECH_TO_TEXT")
|
||||
}
|
||||
if containsStemInTokens(tokens, "фотограф", "фото", "картин", "изображ") && containsStemInTokens(tokens, "прочит", "текст", "надпис") {
|
||||
add("OCR")
|
||||
} else if containsStemInTokens(tokens, "фотограф", "фото", "картин", "изображ") {
|
||||
add("IMAGE_UNDERSTANDING")
|
||||
}
|
||||
if containsStemInTokens(tokens, "видео", "ролик") {
|
||||
add("VIDEO_UNDERSTANDING")
|
||||
}
|
||||
if containsToken(tokens, "напомни") || containsToken(tokens, "напомин") {
|
||||
add("SCHEDULE_REMINDER")
|
||||
}
|
||||
if containsStemInTokens(tokens, "поставь", "создай", "добавь") && containsStemInTokens(tokens, "цел", "задач", "план") {
|
||||
add("COGNITIVE_REASONING")
|
||||
}
|
||||
if isExplicitEmailAction(tokens) {
|
||||
add("SEND_EMAIL")
|
||||
}
|
||||
if containsToken(tokens, "код") || containsToken(tokens, "патч") || containsToken(tokens, "текст") || containsToken(tokens, "главной") || containsToken(tokens, "чате") {
|
||||
if containsToken(tokens, "код") || containsToken(tokens, "патч") || containsToken(tokens, "текст") || containsToken(tokens, "главной") || containsToken(tokens, "чате") || containsToken(tokens, "проект") || containsStemInTokens(tokens, "тест") {
|
||||
if primary == interaction.SpeechActDirective || primary == interaction.SpeechActRequestToInvestigate || secondary == interaction.SpeechActDirective {
|
||||
add("MODIFY_CODE")
|
||||
add("RUN_TESTS")
|
||||
@@ -529,6 +548,9 @@ func capabilityHintsFor(primary, secondary interaction.SpeechAct, tokens []strin
|
||||
if containsInterrogative(tokens) && containsArithmetic(tokens) {
|
||||
add("COGNITIVE_REASONING")
|
||||
}
|
||||
if isCreativeRequestTokens(tokens) && containsStemInTokens(tokens, "погод", "ресторан", "кафе", "билет", "рейс", "цен", "найд", "узнай", "проверь") {
|
||||
add("WEB_RESEARCH")
|
||||
}
|
||||
if containsToken(tokens, "свободно") || containsToken(tokens, "гб") {
|
||||
add("VERIFY_FACT")
|
||||
}
|
||||
@@ -567,8 +589,9 @@ func sourceDependentQuestion(primary, secondary interaction.SpeechAct, tokens []
|
||||
currentCue := containsStemInTokens(tokens, "сейчас", "сегодня", "текущ", "актуаль", "последн", "прямо")
|
||||
observedStateCue := containsStemInTokens(tokens, "погод", "температур", "осад", "ветер", "курс", "цен", "стоим", "новост", "распис", "событ", "статус", "доступен", "работает", "запущен", "срок") || approximateToken(tokens, "погод", 2)
|
||||
placeSearchCue := containsStemInTokens(tokens, "ресторан", "кафе", "отел", "магаз", "мест", "ближ", "адрес", "организац", "бренд", "товар", "артикул", "продавец", "маркетплейс", "wildberries", "wb")
|
||||
travelSearchCue := containsStemInTokens(tokens, "билет", "рейс", "поезд", "автобус", "перелет", "перелёт", "отправлен", "прибыт")
|
||||
return currentCue || observedStateCue || placeSearchCue || travelSearchCue
|
||||
travelSearchCue := containsStemInTokens(tokens, "билет", "рейс", "поезд", "автобус", "перелет", "перелёт", "отправлен", "прибыт", "расстоян", "километр", "добрат", "маршрут")
|
||||
timeDifferenceCue := containsStemInTokens(tokens, "разниц") && containsStemInTokens(tokens, "времен", "час")
|
||||
return currentCue || observedStateCue || placeSearchCue || travelSearchCue || timeDifferenceCue
|
||||
}
|
||||
|
||||
// approximateToken catches ordinary keyboard slips without turning routing
|
||||
|
||||
@@ -0,0 +1,52 @@
|
||||
[
|
||||
{"id":"HH50-001","kind":"question","message":"ты кто?","outcome":"identity"},
|
||||
{"id":"HH50-002","kind":"question","message":"как тебя зовут?","outcome":"identity"},
|
||||
{"id":"HH50-003","kind":"task","message":"я назову тебя Дед, запомни","outcome":"memory"},
|
||||
{"id":"HH50-004","kind":"task","message":"теперь тебя зовут Фея","outcome":"memory"},
|
||||
{"id":"HH50-005","kind":"question","message":"что ты умеешь?","outcome":"conversation"},
|
||||
{"id":"HH50-006","kind":"task","message":"запомни: мне нравится красный цвет","outcome":"memory"},
|
||||
{"id":"HH50-007","kind":"question","message":"что ты помнишь обо мне?","outcome":"memory"},
|
||||
{"id":"HH50-008","kind":"question","message":"выведи мою память таблицей","outcome":"memory"},
|
||||
{"id":"HH50-009","kind":"question","message":"какая сейчас погода в Казани?","outcome":"research"},
|
||||
{"id":"HH50-010","kind":"question","message":"какая разница во времени между Москвой и Шанхаем?","outcome":"research"},
|
||||
{"id":"HH50-011","kind":"assertion","message":"я сейчас в Москве","outcome":"memory"},
|
||||
{"id":"HH50-012","kind":"question","message":"сколько километров до Сочи?","outcome":"research"},
|
||||
{"id":"HH50-013","kind":"question","message":"как можно добраться до Сочи?","outcome":"research"},
|
||||
{"id":"HH50-014","kind":"question","message":"какой ближайший рейс до Казани?","outcome":"research"},
|
||||
{"id":"HH50-015","kind":"question","message":"найди ближайший ресторан в Сочи","outcome":"research"},
|
||||
{"id":"HH50-016","kind":"question","message":"назови доступные модели","outcome":"conversation"},
|
||||
{"id":"HH50-017","kind":"question","message":"какая модель отвечает?","outcome":"conversation"},
|
||||
{"id":"HH50-018","kind":"question","message":"что ты можешь сделать без внешних моделей?","outcome":"conversation"},
|
||||
{"id":"HH50-019","kind":"question","message":"проверь, работает ли интернет","outcome":"research"},
|
||||
{"id":"HH50-020","kind":"question","message":"почему ты не смог найти ответ?","outcome":"conversation"},
|
||||
{"id":"HH50-021","kind":"question","message":"найди на WB бренд marka-wear","outcome":"research"},
|
||||
{"id":"HH50-022","kind":"question","message":"проверь цену на видеокарту RTX 4090","outcome":"research"},
|
||||
{"id":"HH50-023","kind":"question","message":"найди билеты до Ялты","outcome":"research"},
|
||||
{"id":"HH50-024","kind":"question","message":"покажи ближайшие рейсы до Сочи","outcome":"research"},
|
||||
{"id":"HH50-025","kind":"question","message":"какой ближайший самолёт из Москвы в Казань сегодня?","outcome":"research"},
|
||||
{"id":"HH50-026","kind":"question","message":"нарисуй птицу","outcome":"media"},
|
||||
{"id":"HH50-027","kind":"task","message":"нарисуй зелёную птицу","outcome":"media"},
|
||||
{"id":"HH50-028","kind":"task","message":"нарисуй коня","outcome":"media"},
|
||||
{"id":"HH50-029","kind":"task","message":"нарисуй бурого медведя","outcome":"media"},
|
||||
{"id":"HH50-030","kind":"task","message":"нарисуй птицу в HD","outcome":"media"},
|
||||
{"id":"HH50-031","kind":"task","message":"нарисуй круг","outcome":"media"},
|
||||
{"id":"HH50-032","kind":"question","message":"расскажи сказку о Колобке","outcome":"conversation"},
|
||||
{"id":"HH50-033","kind":"task","message":"придумай короткую шутку","outcome":"conversation"},
|
||||
{"id":"HH50-034","kind":"task","message":"исправь опечатку: нарисуй прицу","outcome":"conversation"},
|
||||
{"id":"HH50-035","kind":"task","message":"поставь цель подготовиться к поездке","outcome":"action"},
|
||||
{"id":"HH50-036","kind":"task","message":"напомни завтра в 10 купить билет","outcome":"action"},
|
||||
{"id":"HH50-037","kind":"question","message":"что осталось сделать по цели?","outcome":"action"},
|
||||
{"id":"HH50-038","kind":"task","message":"проверь погоду в Казани и найди рядом кафе","outcome":"multi"},
|
||||
{"id":"HH50-039","kind":"task","message":"узнай погоду в Сочи и нарисуй птицу","outcome":"multi"},
|
||||
{"id":"HH50-040","kind":"task","message":"напомни вечером позвонить и проверь погоду","outcome":"multi"},
|
||||
{"id":"HH50-041","kind":"question","message":"послушай это голосовое сообщение","outcome":"audio"},
|
||||
{"id":"HH50-042","kind":"question","message":"что на этой фотографии?","outcome":"vision"},
|
||||
{"id":"HH50-043","kind":"question","message":"прочитай текст на фотографии","outcome":"ocr"},
|
||||
{"id":"HH50-044","kind":"question","message":"что происходит на этом видео?","outcome":"video"},
|
||||
{"id":"HH50-045","kind":"assertion","message":"источники противоречат друг другу","outcome":"conversation"},
|
||||
{"id":"HH50-046","kind":"question","message":"основной runtime сейчас недоступен","outcome":"research"},
|
||||
{"id":"HH50-047","kind":"task","message":"купи билет до Казани","outcome":"confirmation"},
|
||||
{"id":"HH50-048","kind":"task","message":"отправь письмо о поездке","outcome":"confirmation"},
|
||||
{"id":"HH50-049","kind":"task","message":"измени проект и запусти тесты","outcome":"agent"},
|
||||
{"id":"HH50-050","kind":"question","message":"не получилось ли действие?","outcome":"conversation"}
|
||||
]
|
||||
@@ -0,0 +1,90 @@
|
||||
package household_scenarios
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
"git.cin.su/m/Assistant/contracts/capability"
|
||||
interactiondomain "git.cin.su/m/Assistant/modules/interaction/domain"
|
||||
)
|
||||
|
||||
type checklistItem struct {
|
||||
ID string `json:"id"`
|
||||
Kind string `json:"kind"`
|
||||
Message string `json:"message"`
|
||||
Outcome string `json:"outcome"`
|
||||
}
|
||||
|
||||
func TestHouseholdChecklist50ReachesPositiveSemanticReaction(t *testing.T) {
|
||||
data, err := os.ReadFile(filepath.Join("checklist_50.json"))
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
var items []checklistItem
|
||||
if err := json.Unmarshal(data, &items); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if len(items) != 50 {
|
||||
t.Fatalf("checklist has %d items, want exactly 50", len(items))
|
||||
}
|
||||
seen := map[string]bool{}
|
||||
for _, item := range items {
|
||||
if item.ID == "" || item.Kind == "" || strings.TrimSpace(item.Message) == "" || item.Outcome == "" {
|
||||
t.Fatalf("incomplete checklist item: %+v", item)
|
||||
}
|
||||
if seen[item.ID] {
|
||||
t.Fatalf("duplicate checklist item %q", item.ID)
|
||||
}
|
||||
seen[item.ID] = true
|
||||
checkSemanticReaction(t, item)
|
||||
}
|
||||
}
|
||||
|
||||
func checkSemanticReaction(t *testing.T, item checklistItem) {
|
||||
t.Helper()
|
||||
candidate := interactiondomain.BuildDeterministicCandidate(item.Message, "", time.Unix(10, 0).UnixMilli())
|
||||
has := func(kind capability.CapabilityKind) bool {
|
||||
for _, hint := range candidate.CapabilityHints {
|
||||
if capability.CapabilityKind(hint) == kind {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
switch item.Outcome {
|
||||
case "identity", "memory", "conversation", "confirmation":
|
||||
if has(capability.CapabilityKindWebResearch) || has(capability.CapabilityKindImageGeneration) {
|
||||
t.Errorf("%s %q routed unrelated capability for %s: %+v", item.ID, item.Message, item.Outcome, candidate)
|
||||
}
|
||||
case "research":
|
||||
if !has(capability.CapabilityKindWebResearch) {
|
||||
t.Errorf("%s %q did not select research route: %+v", item.ID, item.Message, candidate)
|
||||
}
|
||||
case "media":
|
||||
if !has(capability.CapabilityKindImageGeneration) {
|
||||
t.Errorf("%s %q did not select image route: %+v", item.ID, item.Message, candidate)
|
||||
}
|
||||
case "action":
|
||||
if !has(capability.CapabilityKindScheduleReminder) && !has(capability.CapabilityKindCognitiveReasoning) && !candidate.NeedsCognition {
|
||||
t.Errorf("%s %q has no action route or cognitive handoff: %+v", item.ID, item.Message, candidate)
|
||||
}
|
||||
case "agent":
|
||||
if !candidate.NeedsCognition && len(candidate.CapabilityHints) == 0 {
|
||||
t.Errorf("%s %q has no agent route or cognitive handoff: %+v", item.ID, item.Message, candidate)
|
||||
}
|
||||
case "multi":
|
||||
if len(candidate.CapabilityHints) < 2 {
|
||||
t.Errorf("%s %q was not decomposed into multiple capabilities: %+v", item.ID, item.Message, candidate)
|
||||
}
|
||||
case "audio", "vision", "ocr", "video":
|
||||
if len(candidate.CapabilityHints) == 0 && !candidate.NeedsCognition {
|
||||
t.Errorf("%s %q has no modality route or cognitive handoff: %+v", item.ID, item.Message, candidate)
|
||||
}
|
||||
default:
|
||||
t.Fatalf("%s uses unknown outcome %q", item.ID, item.Outcome)
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user