Monitoring

1from langtrain import Monitor
2
3# Initialize monitoring
4monitor = Monitor(
5    project_name='my_project',
6    experiment_name='bert_fine_tuning',
7    tracking_uri='http://localhost:5000'
8)
9
10# Start monitoring training
11monitor.start_training(
12    model=model,
13    train_data=train_dataset,
14    val_data=val_dataset,
15    metrics=['loss', 'accuracy', 'f1_score'],
16    log_frequency=100  # Log every 100 steps
17)
18
19# Log custom metrics
20monitor.log_metric('learning_rate', 0.001, step=epoch)
21monitor.log_metric('batch_size', 32)
22monitor.log_artifact('model_config.json', config)

1# Performance monitoring setup
2from langtrain.monitoring import PerformanceMonitor
3
4perf_monitor = PerformanceMonitor(
5    model=model,
6    baseline_metrics={
7        'accuracy': 0.92,
8        'latency_p95': 100,  # milliseconds
9        'throughput': 1000   # requests/second
10    }
11)
12
13# Monitor inference performance
14@perf_monitor.track_inference
15def predict(inputs):
16    return model.predict(inputs)
17
18# Set up alerts for performance degradation
19perf_monitor.set_alert(
20    metric='accuracy',
21    threshold=0.85,
22    comparison='less_than',
23    action='email_alert'
24)
25
26# Generate performance reports
27report = perf_monitor.generate_report(
28    time_range='last_7_days',
29    include_trends=True
30)

1# System resource monitoring
2from langtrain.monitoring import SystemMonitor
3
4sys_monitor = SystemMonitor(
5    track_gpu=True,
6    track_memory=True,
7    track_disk=True,
8    track_network=True
9)
10
11# Start system monitoring
12sys_monitor.start()
13
14# Get current resource usage
15resources = sys_monitor.get_current_usage()
16print(f"GPU Utilization: {resources['gpu_utilization']}%")
17print(f"Memory Usage: {resources['memory_usage']}%")
18print(f"Disk I/O: {resources['disk_io']} MB/s")
19
20# Set resource alerts
21sys_monitor.set_alert(
22    metric='gpu_memory',
23    threshold=90,  # Alert at 90% GPU memory usage
24    action='scale_resources'
25)
26
27# Log resource metrics
28sys_monitor.log_to_dashboard(dashboard_url='http://grafana:3000')

1# Data quality monitoring
2from langtrain.monitoring import DataMonitor
3
4data_monitor = DataMonitor(
5    reference_data=training_data,
6    feature_columns=['text_length', 'sentiment_score'],
7    categorical_columns=['category', 'language']
8)
9
10# Monitor incoming data
11@data_monitor.track_data_quality
12def process_batch(batch_data):
13    # Your data processing logic
14    predictions = model.predict(batch_data)
15    return predictions
16
17# Detect data drift
18drift_report = data_monitor.detect_drift(
19    new_data=production_data,
20    drift_methods=['ks_test', 'chi_square', 'jensen_shannon']
21)
22
23if drift_report.has_drift:
24    print(f"Data drift detected in features: {drift_report.drifted_features}")
25    
26# Set up data quality alerts
27data_monitor.configure_alerts(
28    drift_threshold=0.1,
29    quality_threshold=0.95,
30    notification_channels=['email', 'slack']
31)

1# Custom dashboard creation
2from langtrain.monitoring import Dashboard
3
4dashboard = Dashboard(name='Model Performance Dashboard')
5
6# Add metric widgets
7dashboard.add_widget(
8    type='line_chart',
9    title='Training Loss',
10    metrics=['train_loss', 'val_loss'],
11    time_range='last_24_hours'
12)
13
14dashboard.add_widget(
15    type='gauge',
16    title='Current Accuracy',
17    metric='accuracy',
18    min_value=0.0,
19    max_value=1.0,
20    threshold_ranges=[
21        {'min': 0.0, 'max': 0.7, 'color': 'red'},
22        {'min': 0.7, 'max': 0.85, 'color': 'yellow'},
23        {'min': 0.85, 'max': 1.0, 'color': 'green'}
24    ]
25)
26
27dashboard.add_widget(
28    type='table',
29    title='Model Comparison',
30    data_source='model_comparison_results',
31    columns=['model_name', 'accuracy', 'f1_score', 'latency']
32)
33
34# Deploy dashboard
35dashboard.deploy(url='http://monitoring:8080/dashboard')