Monitoring

from langtrain import Monitor

# Initialize monitoring
monitor = Monitor(
    project_name='my_project',
    experiment_name='bert_fine_tuning',
    tracking_uri='http://localhost:5000'
)

# Start monitoring training
monitor.start_training(
    model=model,
    train_data=train_dataset,
    val_data=val_dataset,
    metrics=['loss', 'accuracy', 'f1_score'],
    log_frequency=100  # Log every 100 steps
)

# Log custom metrics
monitor.log_metric('learning_rate', 0.001, step=epoch)
monitor.log_metric('batch_size', 32)
monitor.log_artifact('model_config.json', config)

# Performance monitoring setup
from langtrain.monitoring import PerformanceMonitor

perf_monitor = PerformanceMonitor(
    model=model,
    baseline_metrics={
        'accuracy': 0.92,
        'latency_p95': 100,  # milliseconds
        'throughput': 1000   # requests/second
    }
)

# Monitor inference performance
@perf_monitor.track_inference
def predict(inputs):
    return model.predict(inputs)

# Set up alerts for performance degradation
perf_monitor.set_alert(
    metric='accuracy',
    threshold=0.85,
    comparison='less_than',
    action='email_alert'
)

# Generate performance reports
report = perf_monitor.generate_report(
    time_range='last_7_days',
    include_trends=True
)

# System resource monitoring
from langtrain.monitoring import SystemMonitor

sys_monitor = SystemMonitor(
    track_gpu=True,
    track_memory=True,
    track_disk=True,
    track_network=True
)

# Start system monitoring
sys_monitor.start()

# Get current resource usage
resources = sys_monitor.get_current_usage()
print(f"GPU Utilization: {resources['gpu_utilization']}%")
print(f"Memory Usage: {resources['memory_usage']}%")
print(f"Disk I/O: {resources['disk_io']} MB/s")

# Set resource alerts
sys_monitor.set_alert(
    metric='gpu_memory',
    threshold=90,  # Alert at 90% GPU memory usage
    action='scale_resources'
)

# Log resource metrics
sys_monitor.log_to_dashboard(dashboard_url='http://grafana:3000')

# Data quality monitoring
from langtrain.monitoring import DataMonitor

data_monitor = DataMonitor(
    reference_data=training_data,
    feature_columns=['text_length', 'sentiment_score'],
    categorical_columns=['category', 'language']
)

# Monitor incoming data
@data_monitor.track_data_quality
def process_batch(batch_data):
    # Your data processing logic
    predictions = model.predict(batch_data)
    return predictions

# Detect data drift
drift_report = data_monitor.detect_drift(
    new_data=production_data,
    drift_methods=['ks_test', 'chi_square', 'jensen_shannon']
)

if drift_report.has_drift:
    print(f"Data drift detected in features: {drift_report.drifted_features}")
    
# Set up data quality alerts
data_monitor.configure_alerts(
    drift_threshold=0.1,
    quality_threshold=0.95,
    notification_channels=['email', 'slack']
)

# Custom dashboard creation
from langtrain.monitoring import Dashboard

dashboard = Dashboard(name='Model Performance Dashboard')

# Add metric widgets
dashboard.add_widget(
    type='line_chart',
    title='Training Loss',
    metrics=['train_loss', 'val_loss'],
    time_range='last_24_hours'
)

dashboard.add_widget(
    type='gauge',
    title='Current Accuracy',
    metric='accuracy',
    min_value=0.0,
    max_value=1.0,
    threshold_ranges=[
        {'min': 0.0, 'max': 0.7, 'color': 'red'},
        {'min': 0.7, 'max': 0.85, 'color': 'yellow'},
        {'min': 0.85, 'max': 1.0, 'color': 'green'}
    ]
)

dashboard.add_widget(
    type='table',
    title='Model Comparison',
    data_source='model_comparison_results',
    columns=['model_name', 'accuracy', 'f1_score', 'latency']
)

# Deploy dashboard
dashboard.deploy(url='http://monitoring:8080/dashboard')