AutoCheckpointingOptions

Options & Configuration · AiDotNet.Models.Options

Configuration options for automatic checkpointing during training, supporting step-frequency, monotonic wall-clock interval, and metric-improvement triggers.

Inheritance: ModelOptions → AutoCheckpointingOptions

For Beginners

Automatic checkpointing saves your model’s training progress periodically so you never lose hours of work if a machine restarts or a long run is interrupted. You can trigger saves:

or any combination of the three.

Constructors

Constructor Summary
AutoCheckpointingOptions() Initializes a new instance of the AutoCheckpointingOptions class with default values.
AutoCheckpointingOptions(AutoCheckpointingOptions other) Initializes a new instance of the AutoCheckpointingOptions class by copying another instance.

Properties

Property Summary
TimeSpan? SaveInterval { get; set; } Gets or sets an optional monotonic wall-clock interval (measured via GetTimestamp) after which the next completed training step triggers an automatic checkpoint.
bool MinimizeMetric { get; set; } Gets or sets whether lower values of MetricName represent improvement (true) or higher values represent improvement (false).
bool SaveOnImprovement { get; set; } Gets or sets whether to save a checkpoint whenever the tracked metric improves.
int KeepLast { get; set; } Gets or sets the number of most recent checkpoints to retain when cleaning up older checkpoints.
int SaveFrequency { get; set; } Gets or sets the number of steps between automatic checkpoint saves.
string MetricName { get; set; } Gets or sets the name of the metric tracked when SaveOnImprovement is enabled.

Methods

Method Summary
void Validate() Validates the configured auto-checkpointing options.