SummaryWriter

Models & Types · AiDotNet.Logging

PyTorch-compatible SummaryWriter for TensorBoard logging.

Implements: IDisposable

For Beginners

This is your interface to TensorBoard visualization.

During training, you use this writer to record:

  • Loss values at each step (add_scalar)
  • Model weight distributions (add_histogram)
  • Sample outputs or feature maps (add_image)
  • Model structure (add_graph)

Then you can visualize all this in TensorBoard by running: tensorboard –logdir=your_log_directory

Example usage:

How It Works

This class provides an API similar to PyTorch’s torch.utils.tensorboard.SummaryWriter, making it easy to log training metrics, model weights, images, and more.

Constructors

Constructor Summary
SummaryWriter(string logDir, string comment, int purgeStep, int maxQueue, int flushSecs, string filename) Creates a new SummaryWriter.

Properties

Property Summary
long DefaultStep { get; } Gets the current default step number.
string LogDir { get; } Gets the log directory path.

Methods

Method Summary
void AddCustomScalar(string tag, float value, long? step) Adds a custom scalar with layout.
void AddEmbedding(string tag, float[] embeddings, string[] metadata, float[] labelImg, long? step) Adds an embedding with optional metadata and labels.
void AddHistogram(string tag, ReadOnlySpan<float> values, long? step, int bins) Adds a histogram from a span of values.
void AddHistogram(string tag, float[] values, long? step, int bins) Adds a histogram from a span of values.
void AddHistogram(string tag, float[] values, long? step, int bins) Adds a histogram from a span of values.
void AddHparams(Dictionary<string, object> hparams, Dictionary<string, float> metrics, Dictionary<string, object[]> hparamDomainDiscrete) Adds hyperparameters and associated metrics.
void AddImage(string tag, float[] imageData, long? step, string dataformats) Adds an image to the summary.
void AddImageRaw(string tag, byte[] pixels, int height, int width, int channels, long? step) Adds an image from raw pixel data.
void AddImages(string tag, float[] images, long? step, int nrow, int padding, bool normalize) Adds a grid of images.
void AddPrCurve(string tag, int[] labels, float[] predictions, long? step, int numThresholds) Adds a PR curve for binary classification evaluation.
void AddScalar(string tag, double value, long? step) Adds a scalar value (double precision).
void AddScalar(string tag, float value, long? step) Adds a scalar value (double precision).
void AddScalars(string mainTag, Dictionary<string, float> tagScalarDict, long? step) Adds multiple scalars under a main tag.
void AddText(string tag, string text, long? step) Adds text to the summary.
void Close() Closes the writer (alias for Dispose).
void Dispose() Releases resources and closes the writer.
void Flush() Flushes pending writes to disk.
void LogTrainingStep(float loss, float? accuracy, float? learningRate, long? step) Logs training metrics at the current step.
void LogValidationStep(float loss, float? accuracy, long? step) Logs validation metrics.
void LogWeights(string layerName, float[] weights, float[] gradients, long? step) Logs model weight statistics.