EarlyStopping<T>
Models & Types · AiDotNet.HyperparameterOptimization
Provides early stopping functionality for hyperparameter optimization and training.
How It Works
For Beginners: Early stopping is a technique to:
- Stop training when performance stops improving
- Prevent overfitting by not training too long
- Save compute resources by terminating hopeless trials
Key concepts:
- Patience: How many checks without improvement before stopping
- Min Delta: Minimum improvement to count as “better”
- Best Value: The best score seen so far
- Counter: Tracks consecutive non-improvements
Early stopping is essential for efficient hyperparameter search because it allows quick termination of poor configurations.
Constructors
| Constructor | Summary |
|---|---|
EarlyStopping(int patience, double minDelta, bool maximize, EarlyStoppingMode mode) |
Initializes a new instance of the EarlyStopping class. |
Properties
| Property | Summary |
|---|---|
IReadOnlyList<double> History { get; } |
Gets the history of values that were checked. |
bool ShouldStop { get; } |
Gets whether early stopping has been triggered. |
double BestValue { get; } |
Gets the best value observed. |
int BestEpoch { get; } |
Gets the epoch at which the best value was observed. |
int EpochsSinceBest { get; } |
Gets the number of epochs since the best value was observed. |
Methods
| Method | Summary |
|---|---|
EarlyStoppingState GetState() |
Gets a summary of the early stopping state. |
bool Check(T value, int epoch) |
Checks if training should stop based on the new value. |
bool Check(double value, int epoch) |
Checks if training should stop based on the new value. |
void Reset() |
Resets the early stopping state. |