Interface IExperimentTracker<T>

Namespace

AiDotNet.Interfaces

Assembly

AiDotNet.dll

Defines the contract for experiment tracking systems that log machine learning experiments.

public interface IExperimentTracker<T>

Type Parameters

T

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

Remarks

An experiment tracker records information about machine learning experiments including parameters, metrics, and artifacts to enable reproducibility and comparison of different training runs.

For Beginners: Think of an experiment tracker as a lab notebook for machine learning. Just like a scientist records their experimental conditions and results, an experiment tracker logs all the details of your machine learning model training - what settings you used, how well it performed, and what models you created.

Key capabilities include:

• Creating and managing experiments (groups of related training runs)
• Logging hyperparameters (settings used for training)
• Recording metrics (performance measurements over time)
• Storing artifacts (models, plots, data files)
• Comparing different training runs
• Reproducing previous experiments

Why experiment tracking matters:

• Helps you keep track of what you've tried
• Makes it easy to reproduce good results
• Enables comparison between different approaches
• Provides audit trail for model development

Methods

CreateExperiment(string, string?, Dictionary<string, string>?)

Creates a new experiment to organize related training runs.

string CreateExperiment(string name, string? description = null, Dictionary<string, string>? tags = null)

Parameters

name string

The name of the experiment.

description string

Optional description of the experiment's purpose.

tags Dictionary<string, string>

Optional tags to categorize the experiment.

Returns

string

The unique identifier for the created experiment.

Remarks

For Beginners: An experiment is a collection of related training runs. For example, you might create an experiment called "House Price Prediction" and then have multiple runs within it trying different algorithms or settings.

DeleteExperiment(string)

Deletes an experiment and all its associated runs.

void DeleteExperiment(string experimentId)

Parameters

experimentId string

The ID of the experiment to delete.

DeleteRun(string)

Deletes a specific run.

void DeleteRun(string runId)

Parameters

runId string

The ID of the run to delete.

GetExperiment(string)

Gets an existing experiment by its ID.

IExperiment GetExperiment(string experimentId)

Parameters

experimentId string

The unique identifier of the experiment.

Returns

IExperiment

An IExperiment object containing experiment details.

GetRun(string)

Gets an existing run by its ID.

IExperimentRun<T> GetRun(string runId)

Parameters

runId string

The unique identifier of the run.

Returns

IExperimentRun<T>

An IExperimentRun object containing run details.

ListExperiments(string?)

Lists all experiments, optionally filtered by criteria.

IEnumerable<IExperiment> ListExperiments(string? filter = null)

Parameters

filter string

Optional filter expression.

Returns

IEnumerable<IExperiment>

A list of experiments matching the criteria.

ListRuns(string, string?)

Lists all runs in an experiment, optionally filtered by criteria.

IEnumerable<IExperimentRun<T>> ListRuns(string experimentId, string? filter = null)

Parameters

experimentId string

The experiment ID to list runs from.

filter string

Optional filter expression.

Returns

IEnumerable<IExperimentRun<T>>

A list of runs matching the criteria.

SearchRuns(string, int)

Searches for runs across all experiments based on criteria.

IEnumerable<IExperimentRun<T>> SearchRuns(string filter, int maxResults = 100)

Parameters

filter string

Filter expression for searching runs.

maxResults int

Maximum number of results to return.

Returns

IEnumerable<IExperimentRun<T>>

A list of runs matching the search criteria.

StartRun(string, string?, Dictionary<string, string>?)

Starts a new training run within an experiment.

IExperimentRun<T> StartRun(string experimentId, string? runName = null, Dictionary<string, string>? tags = null)

Parameters

experimentId string

The ID of the experiment this run belongs to.

runName string

Optional name for this specific run.

tags Dictionary<string, string>

Optional tags to categorize the run.

Returns

IExperimentRun<T>

An IExperimentRun object for logging metrics and parameters.

Remarks

For Beginners: A run represents a single training session with specific settings. Each time you train a model with different parameters, you start a new run.