Zum Hauptinhalt springen
Unkalibrierte Beta
StrikeStrike

Probabilities in beta

How to use strike-1 probabilities while Strike is in beta.

Beta

strike-1 gives a probability on every answer. During beta, use those numbers to compare options and route work, not as forecasts or risk scores. We have not published accuracy benchmarks yet.

Response fields

Responses may include the fields calibrated (boolean) and calibration_version. In beta, calibrated is typically false.

Python
import strikeflow as strike

result = strike.decide("Is this spam?", "yes_no", context="WIN A FREE IPHONE!!! click now")
print(result.calibrated, result.calibration_version)

Log these fields; do not show them to your end users as accuracy claims.

The Changelog lists changes to how probabilities are produced.

debias option

StrikeModel(debias=True) sends options.debias = "permute" to the API for choice decisions. This is experimental and reduces variance from option order.

Python
import strikeflow as strike

engine = strike.Strike(debias=True)
result = engine.decide("Pick one", ["a", "b"], context="trivial")
Probabilities in beta: Strike docs