Class MetaTrainingStepResult<T>

Namespace

AiDotNet.Models.Results

Assembly

AiDotNet.dll

Results from a single meta-training step (one outer loop update).

public class MetaTrainingStepResult<T>

Type Parameters

T

The numeric type used for calculations (e.g., float, double, decimal).

Inheritance

object

MetaTrainingStepResult<T>

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

Remarks

This class represents metrics from one iteration of meta-training, tracking the performance of a single outer loop update across a batch of tasks. It's a lightweight snapshot designed for real-time monitoring during training.

For Beginners: Think of this as the "score" for one training iteration.

Each meta-training iteration:

1. Samples a batch of tasks (e.g., 4 tasks)
2. Adapts to each task (inner loop)
3. Updates meta-parameters based on adaptation results (outer loop)
4. Returns these metrics to show how well that update performed

You'll get one of these for each iteration during training, allowing you to:

• Monitor training progress in real-time
• Log metrics to TensorBoard or similar tools
• Implement early stopping or learning rate schedules
• Debug training issues as they occur

Constructors

MetaTrainingStepResult(T, T, T, int, int, double, Dictionary<string, T>?)

Initializes a new instance with metrics from one meta-training step.

public MetaTrainingStepResult(T metaLoss, T taskLoss, T accuracy, int numTasks, int iteration, double timeMs, Dictionary<string, T>? additionalMetrics = null)

Parameters

metaLoss T

The meta-loss for this step.

taskLoss T

The average task-specific loss.

accuracy T

The average accuracy across tasks.

numTasks int

The number of tasks in the meta-batch.

iteration int

The iteration number.

timeMs double

Time taken for this step in milliseconds.

additionalMetrics Dictionary<string, T>

Optional algorithm-specific metrics.

Remarks

For Beginners: Call this after each meta-training step to package the results. This makes it easy to log, visualize, and monitor training progress.

Properties

Accuracy

Gets the average accuracy across tasks in this step.

public T Accuracy { get; }

Property Value

T

The mean accuracy on query sets after adaptation, across all tasks in the batch.

AdditionalMetrics

Gets algorithm-specific metrics for this training step.

public Dictionary<string, T> AdditionalMetrics { get; }

Property Value

Dictionary<string, T>

A dictionary of custom metrics with generic T values.

Remarks

For Production: Common additional metrics include: - "gradient_norm": Magnitude of meta-gradients - "support_accuracy": Accuracy on support sets (should be very high) - "learning_rate": Current learning rate (if using schedules) - "parameter_norm": Magnitude of model parameters

Iteration

Gets the iteration number for this training step.

public int Iteration { get; }

Property Value

int

The sequential iteration count, useful for time-series analysis of training progress.

MetaLoss

Gets the meta-loss for this training step.

public T MetaLoss { get; }

Property Value

T

The loss computed across all tasks in the meta-batch after inner loop adaptation. This is what the outer loop optimizes.

Remarks

For Beginners: This is the main number the meta-learner tries to minimize. Lower meta-loss means the model is getting better at adapting to new tasks quickly.

NumTasks

Gets the number of tasks processed in this meta-training step.

public int NumTasks { get; }

Property Value

int

The meta-batch size for this iteration.

TaskLoss

Gets the average task-specific loss after adaptation.

public T TaskLoss { get; }

Property Value

T

The mean loss across all tasks in the batch, measured on query sets after inner loop adaptation.

TimeMs

Gets the time taken for this meta-training step in milliseconds.

public double TimeMs { get; }

Property Value

double

The elapsed time for completing this outer loop update, including all inner loop adaptations.

Methods

ToString()

Returns a concise string representation for logging.

public override string ToString()

Returns

string