Examples

# models/fraud-detector.model.yaml
model: fraud-detector
version: "1.0.0"
runtime: python3.11

input:
  type: object
  properties:
    transaction_id:
      type: string
      format: uuid
    amount:
      type: number
      minimum: 0
    merchant_category:
      type: string
      enum: [retail, electronics, travel, food, services]
    user_location:
      type: object
      properties:
        lat:
          type: number
        lng:
          type: number
    device_fingerprint:
      type: string
    time_of_day:
      type: number
      minimum: 0
      maximum: 24

output:
  type: object
  properties:
    is_fraud:
      type: boolean
    confidence:
      type: number
      minimum: 0
      maximum: 1
    risk_factors:
      type: array
      items:
        type: string
    recommended_action:
      type: string
      enum: [approve, review, decline]

training:
  datasource: transactions-db
  snapshot: required
  schedule: "0 2 * * *"
  validation:
    split: 0.2
    metrics:
      - accuracy
      - precision
      - recall
      - f1_score
      - auc_roc

inference:
  timeout: 50ms
  retries: 2
  cache:
    enabled: true
    ttl: 0  # No caching for fraud detection
  
drift:
  enabled: true
  baseline: training
  threshold: 0.1
  metrics:
    - feature_distribution
    - prediction_distribution
  action: notify

# datasources/transactions-db.datasource.yaml
datasource: transactions-db
type: postgresql

connection:
  host: ${POSTGRES_HOST}
  port: 5432
  database: fraud_detection
  user: ${POSTGRES_USER}
  password: ${POSTGRES_PASSWORD}
  ssl: required

schema:
  table: transactions
  columns:
    - name: id
      type: uuid
      primary: true
    - name: amount
      type: decimal
    - name: merchant_category
      type: string
    - name: user_lat
      type: decimal
    - name: user_lng
      type: decimal
    - name: device_fingerprint
      type: string
    - name: created_at
      type: timestamp
    - name: is_fraud
      type: boolean
      label: true  # This is the training label

query:
  filter: "created_at >= NOW() - INTERVAL '90 days'"
  
snapshot:
  strategy: incremental
  column: created_at
  retention: 180d

# training/train.py
import pulse
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.preprocessing import StandardScaler
import joblib

def train(data, config):
    """Train fraud detection model."""
    
    # Prepare features
    X = data[['amount', 'merchant_category_encoded', 
              'user_lat', 'user_lng', 'hour_of_day']]
    y = data['is_fraud']
    
    # Scale features
    scaler = StandardScaler()
    X_scaled = scaler.fit_transform(X)
    
    # Train model
    model = GradientBoostingClassifier(
        n_estimators=100,
        max_depth=6,
        learning_rate=0.1,
        random_state=42
    )
    model.fit(X_scaled, y)
    
    # Save artifacts
    pulse.save_artifact('model.joblib', model)
    pulse.save_artifact('scaler.joblib', scaler)
    
    # Log metrics
    pulse.log_metric('accuracy', model.score(X_scaled, y))
    
    return model

def infer(input_data, artifacts):
    """Run inference on new transaction."""
    
    model = artifacts['model.joblib']
    scaler = artifacts['scaler.joblib']
    
    # Prepare features
    features = prepare_features(input_data)
    features_scaled = scaler.transform([features])
    
    # Predict
    proba = model.predict_proba(features_scaled)[0]
    is_fraud = proba[1] > 0.5
    
    # Get risk factors
    risk_factors = explain_prediction(model, features_scaled)
    
    return {
        'is_fraud': is_fraud,
        'confidence': float(proba[1]),
        'risk_factors': risk_factors,
        'recommended_action': get_action(proba[1])
    }

# Validate configuration
pulse validate

# Create initial snapshot
pulse snapshot transactions-db

# Train the model
pulse train fraud-detector --watch

# Deploy to production
pulse deploy fraud-detector

# Test inference
curl -X POST https://api.pulse.dev/v1/infer/fraud-detector \
  -H "Authorization: Bearer $PULSE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "transaction_id": "550e8400-e29b-41d4-a716-446655440000",
    "amount": 1500.00,
    "merchant_category": "electronics",
    "user_location": {"lat": 40.7128, "lng": -74.0060},
    "device_fingerprint": "abc123",
    "time_of_day": 14.5
  }'

# Response
{
  "is_fraud": false,
  "confidence": 0.12,
  "risk_factors": [],
  "recommended_action": "approve"
}