ProfilingConfig

Options & Configuration · AiDotNet.Deployment.Configuration

Configuration for performance profiling during model training and inference.

For Beginners

Profiling measures how long different parts of your ML code take to run. Think of it like a stopwatch for your code - it helps you find bottlenecks and optimize performance.

What gets tracked:

  • Operation timing: How long each training step, forward pass, backward pass takes
  • Memory allocations: How much memory is used during training
  • Call hierarchy: Which operations call which other operations
  • Percentiles: P50 (median), P95, P99 timing for statistical analysis

Why it’s important:

  • Find bottlenecks in your training pipeline
  • Compare performance before and after optimizations
  • Identify memory-intensive operations
  • Understand your model’s computational profile

The profiler uses production-ready algorithms:

  • Welford’s algorithm for O(1) streaming mean/variance
  • Reservoir sampling for bounded-memory percentile estimation
  • Configurable sampling for high-frequency operations

Constructors

Constructor Summary
ProfilingConfig()

Properties

Property Summary
Dictionary<string, string> CustomTags { get; set; } Gets or sets custom tags to include with profiling data.
bool AutoEnableInDebug { get; set; } Gets or sets whether to auto-enable profiling in debug builds (default: true).
bool DetailedTiming { get; set; } Gets or sets whether to include detailed timing breakdowns (default: false).
bool Enabled { get; set; } Gets or sets whether profiling is enabled (default: false).
bool MeasureMemory { get; set; } Alias for TrackAllocations for more intuitive access.
bool TraceExecution { get; set; } Alias for TrackCallHierarchy for more intuitive access.
bool TrackAllocations { get; set; } Gets or sets whether to track memory allocations (default: true).
bool TrackCallHierarchy { get; set; } Gets or sets whether to track parent-child call relationships (default: true).
double SamplingRate { get; set; } Gets or sets the sampling rate for high-frequency operations (default: 1.0 = 100%).
int MaxOperations { get; set; } Gets or sets the maximum number of unique operations to track (default: 10000).
int ReservoirSize { get; set; } Gets or sets the reservoir size for percentile estimation (default: 1000).