Try several sets of worked examples and keep the one that scores best on
the validation rows: none, labeled only, bootstrapped, and bootstrapped
from shuffled rows with random sizes, each scored on valset (default: the
rows). Python's BootstrapFewShotWithRandomSearch. Returns the best copy
and every candidate ({ candidate, demos, score }).
Try several sets of worked examples and keep the one that scores best on the validation rows: none, labeled only, bootstrapped, and bootstrapped from shuffled rows with random sizes, each scored on
valset(default: the rows). Python'sBootstrapFewShotWithRandomSearch. Returns the best copy and every candidate ({ candidate, demos, score }).