Label Training Data

Label Training Data

See Extended Examples for complete config files + templates.

Systematically label data for supervised ML w/ QC + efficient workflows.

Use When

• Start supervised ML needing labeled data
• Model perf limited by insufficient examples
• Label text, images, audio, video
• Measure + improve annotation quality
• Team of annotators w/ diff expertise
• Active learning → prioritize valuable examples
• Track progress + costs
• Consistent labels across multi annotators

In

• Req: Unlabeled dataset (images, text, audio, video)
• Req: Label schema (classes, attributes, annotation types)
• Req: Labeling guidelines doc
• Opt: Pre-existing labels (quality compare)
• Opt: Model predictions for pre-annotation
• Opt: Budget + timeline
• Opt: Domain expert availability

Do

Step 1: Install + Config Label Studio

Setup labeling platform.

# Install Label Studio
pip install label-studio

# Or use Docker for production
docker pull heartexlabs/label-studio:latest

# Create project directory
mkdir -p labeling-project/{data,exports,config}
cd labeling-project

# Initialize Label Studio
label-studio init my_project

# Start Label Studio server
label-studio start my_project --port 8080

Access http://localhost:8080 (default creds: create on first visit).

Prod deploy w/ Docker:

# docker-compose.yml
version: '3.8'

services:
  label-studio:
    image: heartexlabs/label-studio:latest
    ports:
      - "8080:8080"
# ... (see EXAMPLES.md for complete implementation)

docker-compose up -d

→ Label Studio running + accessible, PostgreSQL DB init for prod.

If err: Port 8080 busy → change port in config. Docker fails → check daemon running. Ensure disk space for data vols. Firewall allows 8080.

Step 2: Design Interface + Schema

Create labeling config for task type.

# labeling-project/config/labeling_config.py
"""
Label Studio configuration templates for common tasks.
"""

# Text Classification (single label)
TEXT_CLASSIFICATION = """
<View>
# ... (see EXAMPLES.md for complete implementation)

→ Interface configured w/ appropriate controls for task type, data imported, interface accessible to annotators.

If err: Validate XML config w/ Label Studio validator. Check data file format (JSON / CSV). Ensure image/audio URLs accessible if external storage. Verify API key perms.

Step 3: Prepare Data + Sampling Strategy

Format data for import + prioritize for labeling.

# labeling-project/prepare_data.py
import pandas as pd
import json
import random
from typing import List, Dict
from sklearn.cluster import KMeans
import numpy as np

# ... (see EXAMPLES.md for complete implementation)

→ Data formatted for Label Studio import, sampling prioritizes informative examples, tasks include metadata for tracking.

If err: Verify JSON w/ jq / Py json.load(). Check URLs accessible if remote images. Ensure no special chars break JSON. Validate column names match config.

Step 4: QC + IAA Measurement

Setup processes to measure + improve quality.

# labeling-project/quality_control.py
import pandas as pd
import numpy as np
from sklearn.metrics import cohen_kappa_score, confusion_matrix
from typing import Dict, List, Tuple
import logging

logging.basicConfig(level=logging.INFO)
# ... (see EXAMPLES.md for complete implementation)

→ IAA measured (Cohen's Kappa > 0.6 = moderate, >0.8 = good), difficult tasks ID'd for review, annotator perf tracked.

If err: Kappa very low (<0.4) → review guidelines for clarity, retrain annotators, simplify schema, check ambiguous examples, consider expert annotators for gold std.

Step 5: Export + Integrate Labeled Data

Export labels + prep for ML training.

# labeling-project/export_labels.py
import requests
import pandas as pd
import json
from typing import List, Dict
import logging

logger = logging.getLogger(__name__)
# ... (see EXAMPLES.md for complete implementation)

→ Annotations exported training-ready format, label distribution balanced / documented, data quality validated before training.

If err: Verify API key perms. Check export format compat w/ ML framework. Handle missing annotations gracefully. Validate JSON matches expected.

Step 6: Continuous Labeling Pipeline

Automate workflow w/ active learning.

# labeling-project/active_learning_pipeline.py
import schedule
import time
import logging
from datetime import datetime
from prepare_data import DataSampler, prepare_label_studio_format
from export_labels import LabelStudioExporter, convert_to_training_format
import pandas as pd
# ... (see EXAMPLES.md for complete implementation)

→ Active learning selects informative examples auto, batches prep weekly, model retrained when sufficient new labels avail.

If err: Uncertainty sampling doesn't improve model → try diversity sampling. Annotators can't keep up → reduce batch size. Monitor queue length, backpressure if queue grows.

Check

• Label Studio accessible + responsive
• Interface intuitive (test w/ sample annotator)
• Data import successful w/ correct format
• IAA (Cohen's Kappa) > 0.6
• QC IDs problematic tasks
• Labels export training-ready
• Distribution matches expected (or intentionally imbalanced)
• Active learning pipeline runs w/o manual intervention
• Throughput meets timeline

Traps

• Unclear guidelines: Ambiguous → inconsistent labels. Invest detailed guidelines + examples.
• Insufficient overlap: Can't measure IAA w/o multi annotators per task. 10-20% overlap.
• Ignore difficult cases: Edge cases often skipped, critical for robustness. Flag for expert review.
• Batch effects: Annotator fatigue / learning → temporal inconsistency. Randomize task order.
• No quality feedback: Annotators don't improve w/o feedback. Regular accuracy reports.
• Wrong sampling: Random wastes budget on easy. Use uncertainty / diversity sampling.
• Labeling in isolation: Domain experts needed for complex tasks. Pair novices w/ experts initially.
• Not tracking costs: Labeling expensive. Monitor time per task + budget consumption.

→

• version-ml-data — version control for labeled datasets
• track-ml-experiments — track model perf as labels added