Ai Annotation Workflow

by dengineproblem

workflowdata

Эксперт по data annotation. Используй для ML labeling, annotation workflows и quality control.

Skill Details

Repository Files

1 file in this skill directory


name: ai-annotation-workflow description: Эксперт по data annotation. Используй для ML labeling, annotation workflows и quality control.

AI Annotation Workflow Expert

Эксперт по проектированию и управлению процессами разметки данных для машинного обучения.

Основные принципы

Цикл разметки данных

  1. Планирование: Определение задачи, создание guidelines
  2. Пилот: Тестирование с малой выборкой
  3. Масштабирование: Расширение на полный датасет
  4. Контроль качества: Постоянный мониторинг
  5. Итерация: Улучшение на основе обратной связи

Типы задач разметки

  • Классификация: Категоризация данных
  • Детекция объектов: Bounding boxes
  • Сегментация: Пиксельная разметка
  • NER: Named Entity Recognition
  • Sentiment: Анализ тональности
  • Транскрипция: Аудио в текст

Создание Guidelines

Структура документа

# Annotation Guidelines v1.0

## Задача
Разметка изображений товаров для e-commerce

## Категории
1. Одежда
   - Верхняя одежда
   - Нижняя одежда
   - Аксессуары

2. Электроника
   - Смартфоны
   - Ноутбуки
   - Аксессуары

## Правила разметки
- Выбирайте наиболее специфичную категорию
- При неопределённости используйте "Другое"
- Один объект = одна категория

## Примеры
[Примеры с изображениями и правильной разметкой]

## Edge Cases
- Товар частично виден: размечайте если >50% видно
- Несколько товаров: размечайте каждый отдельно

Метрики качества

Inter-Annotator Agreement (IAA)

from sklearn.metrics import cohen_kappa_score, fleiss_kappa

def calculate_agreement(annotations_a, annotations_b):
    """Расчёт согласованности между аннотаторами"""

    # Cohen's Kappa для двух аннотаторов
    kappa = cohen_kappa_score(annotations_a, annotations_b)

    # Интерпретация
    if kappa < 0.20:
        interpretation = "Poor"
    elif kappa < 0.40:
        interpretation = "Fair"
    elif kappa < 0.60:
        interpretation = "Moderate"
    elif kappa < 0.80:
        interpretation = "Good"
    else:
        interpretation = "Excellent"

    return {
        'kappa': kappa,
        'interpretation': interpretation
    }

Quality Metrics

class AnnotationQualityMonitor:
    def __init__(self):
        self.metrics = []

    def calculate_metrics(self, annotations, gold_standard):
        """Расчёт метрик качества относительно эталона"""

        from sklearn.metrics import precision_score, recall_score, f1_score

        precision = precision_score(gold_standard, annotations, average='weighted')
        recall = recall_score(gold_standard, annotations, average='weighted')
        f1 = f1_score(gold_standard, annotations, average='weighted')

        return {
            'precision': precision,
            'recall': recall,
            'f1': f1,
            'accuracy': sum(a == g for a, g in zip(annotations, gold_standard)) / len(annotations)
        }

    def detect_drift(self, window_size=100):
        """Обнаружение дрифта качества"""

        if len(self.metrics) < window_size * 2:
            return False

        recent = self.metrics[-window_size:]
        previous = self.metrics[-window_size*2:-window_size]

        recent_avg = sum(m['f1'] for m in recent) / len(recent)
        previous_avg = sum(m['f1'] for m in previous) / len(previous)

        # Дрифт если падение > 5%
        return (previous_avg - recent_avg) / previous_avg > 0.05

Workflow автоматизации

Label Studio интеграция

from label_studio_sdk import Client

class AnnotationPipeline:
    def __init__(self, api_key, url):
        self.client = Client(url=url, api_key=api_key)

    def create_project(self, name, label_config):
        """Создание проекта разметки"""

        project = self.client.create_project(
            title=name,
            label_config=label_config
        )
        return project

    def import_tasks(self, project_id, data):
        """Импорт задач для разметки"""

        project = self.client.get_project(project_id)
        project.import_tasks(data)

    def export_annotations(self, project_id, format='JSON'):
        """Экспорт готовых аннотаций"""

        project = self.client.get_project(project_id)
        return project.export_tasks(format=format)

Active Learning Pipeline

class ActiveLearningAnnotation:
    def __init__(self, model, unlabeled_pool):
        self.model = model
        self.unlabeled_pool = unlabeled_pool
        self.labeled_data = []

    def select_samples_for_annotation(self, n_samples=100, strategy='uncertainty'):
        """Выбор образцов для разметки"""

        if strategy == 'uncertainty':
            # Выбор образцов с наибольшей неопределённостью
            predictions = self.model.predict_proba(self.unlabeled_pool)
            uncertainties = -np.sum(predictions * np.log(predictions + 1e-10), axis=1)
            indices = np.argsort(uncertainties)[-n_samples:]

        elif strategy == 'diversity':
            # Выбор разнообразных образцов
            from sklearn.cluster import KMeans
            kmeans = KMeans(n_clusters=n_samples)
            kmeans.fit(self.unlabeled_pool)
            indices = [
                np.argmin(np.linalg.norm(self.unlabeled_pool - center, axis=1))
                for center in kmeans.cluster_centers_
            ]

        return self.unlabeled_pool[indices]

    def update_model(self, new_annotations):
        """Обновление модели после получения аннотаций"""

        self.labeled_data.extend(new_annotations)
        X = [item['features'] for item in self.labeled_data]
        y = [item['label'] for item in self.labeled_data]
        self.model.fit(X, y)

Управление аннотаторами

Онбординг

class AnnotatorOnboarding:
    def __init__(self, gold_standard_samples):
        self.gold_standard = gold_standard_samples
        self.passing_threshold = 0.85

    def run_qualification_test(self, annotator_id, annotations):
        """Квалификационный тест для новых аннотаторов"""

        correct = sum(
            a == g for a, g in zip(annotations, self.gold_standard)
        )
        accuracy = correct / len(self.gold_standard)

        return {
            'annotator_id': annotator_id,
            'accuracy': accuracy,
            'passed': accuracy >= self.passing_threshold,
            'errors': [
                {'index': i, 'expected': g, 'actual': a}
                for i, (a, g) in enumerate(zip(annotations, self.gold_standard))
                if a != g
            ]
        }

Мониторинг производительности

class AnnotatorPerformanceTracker:
    def __init__(self):
        self.annotator_stats = {}

    def log_annotation(self, annotator_id, task_id, time_spent, quality_score):
        if annotator_id not in self.annotator_stats:
            self.annotator_stats[annotator_id] = []

        self.annotator_stats[annotator_id].append({
            'task_id': task_id,
            'time_spent': time_spent,
            'quality_score': quality_score,
            'timestamp': datetime.now()
        })

    def get_annotator_report(self, annotator_id):
        stats = self.annotator_stats.get(annotator_id, [])

        if not stats:
            return None

        return {
            'total_tasks': len(stats),
            'avg_time': sum(s['time_spent'] for s in stats) / len(stats),
            'avg_quality': sum(s['quality_score'] for s in stats) / len(stats),
            'tasks_per_hour': len(stats) / (
                (stats[-1]['timestamp'] - stats[0]['timestamp']).total_seconds() / 3600
            ) if len(stats) > 1 else 0
        }

Инструменты разметки

Популярные платформы

  • Label Studio: Open-source, гибкая конфигурация
  • Labelbox: Enterprise, ML-assisted labeling
  • Scale AI: Managed workforce
  • Amazon SageMaker Ground Truth: AWS интеграция
  • Prodigy: NLP-focused, active learning

Выбор инструмента

Критерий Label Studio Labelbox Scale AI
Стоимость Free/Open $$ $$$
Workforce Self-managed Optional Included
ML Assist Basic Advanced Advanced
Customization High Medium Low

Лучшие практики

  1. Начинайте с пилота — 100-200 образцов для калибровки
  2. Итерируйте guidelines — обновляйте по мере обнаружения edge cases
  3. Используйте gold standard — 5-10% данных для контроля качества
  4. Балансируйте скорость и качество — не давите на аннотаторов
  5. Документируйте решения — записывайте все разъяснения
  6. Автоматизируйте что можно — pre-labeling, validation rules

Related Skills

Xlsx

Comprehensive spreadsheet creation, editing, and analysis with support for formulas, formatting, data analysis, and visualization. When Claude needs to work with spreadsheets (.xlsx, .xlsm, .csv, .tsv, etc) for: (1) Creating new spreadsheets with formulas and formatting, (2) Reading or analyzing data, (3) Modify existing spreadsheets while preserving formulas, (4) Data analysis and visualization in spreadsheets, or (5) Recalculating formulas

data

Clickhouse Io

ClickHouse database patterns, query optimization, analytics, and data engineering best practices for high-performance analytical workloads.

datacli

Clickhouse Io

ClickHouse database patterns, query optimization, analytics, and data engineering best practices for high-performance analytical workloads.

datacli

Analyzing Financial Statements

This skill calculates key financial ratios and metrics from financial statement data for investment analysis

data

Data Storytelling

Transform data into compelling narratives using visualization, context, and persuasive structure. Use when presenting analytics to stakeholders, creating data reports, or building executive presentations.

data

Kpi Dashboard Design

Design effective KPI dashboards with metrics selection, visualization best practices, and real-time monitoring patterns. Use when building business dashboards, selecting metrics, or designing data visualization layouts.

designdata

Dbt Transformation Patterns

Master dbt (data build tool) for analytics engineering with model organization, testing, documentation, and incremental strategies. Use when building data transformations, creating data models, or implementing analytics engineering best practices.

testingdocumenttool

Sql Optimization Patterns

Master SQL query optimization, indexing strategies, and EXPLAIN analysis to dramatically improve database performance and eliminate slow queries. Use when debugging slow queries, designing database schemas, or optimizing application performance.

designdata

Anndata

This skill should be used when working with annotated data matrices in Python, particularly for single-cell genomics analysis, managing experimental measurements with metadata, or handling large-scale biological datasets. Use when tasks involve AnnData objects, h5ad files, single-cell RNA-seq data, or integration with scanpy/scverse tools.

arttooldata

Xlsx

Spreadsheet toolkit (.xlsx/.csv). Create/edit with formulas/formatting, analyze data, visualization, recalculate formulas, for spreadsheet processing and analysis.

tooldata

Skill Information

Category:Data
Last Updated:12/10/2025