DiscreteSearchOptimizer<T>
Models & Types · AiDotNet.PromptEngineering.Optimization
Optimizer that uses discrete search to find better prompts by testing variations.
Inheritance: PromptOptimizerBase<T> → DiscreteSearchOptimizer<T>
For Beginners
Tries different versions of prompt parts and picks what works best.
Example:
var optimizer = new DiscreteSearchOptimizer<double>();
// your labelled examples, and whatever answers a prompt
var testCases = new List<(string Input, string Expected)>
{
("I loved it", "positive"),
("Total waste of time", "negative")
};
Func<string, string> model = prompt => "positive";
// Evaluation function that returns accuracy
double Evaluate(string prompt)
{
var correct = 0;
foreach (var test in testCases)
{
var result = model(prompt + test.Input);
if (result == test.Expected) correct++;
}
return correct / (double)testCases.Count;
}
var optimized = optimizer.Optimize(
initialPrompt: "Classify the sentiment:",
evaluationFunction: Evaluate,
maxIterations: 50
);
// Returns best-performing prompt variation
How It Works
This optimizer generates variations of prompt components and tests combinations to find the best-performing prompt. It’s interpretable and systematic.
Constructors
| Constructor | Summary |
|---|---|
DiscreteSearchOptimizer() |
Initializes a new instance of the DiscreteSearchOptimizer class with default variations. |
DiscreteSearchOptimizer(IEnumerable<string> instructionVariations, IEnumerable<string> formatVariations) |
Initializes a new instance of the DiscreteSearchOptimizer class with custom variations. |
Methods
| Method | Summary |
|---|---|
void AddFormatVariations(string[] variations) |
Adds custom format variations to try during optimization. |
void AddInstructionVariations(string[] variations) |
Adds custom instruction variations to try during optimization. |