Added the Model Comparison assistant for blind comparisons of two models (#1009)

Co-authored-by: Thorsten Sommer <SommerEngineering@users.noreply.github.com>
This commit is contained in:
Dominic NeuburgandThorsten Sommer authored and GitHub committed 2026-11-14 14:11:34 +01:00
1 parent 12f26445ba
commit b7255694cd
221 files changed
+19989 -90

No files matched your search

@@ -0,0 +1,71 @@
using AIStudio.Assistants.ModelComparison;
namespace AIStudio.Tests.Assistants.ModelComparison;
/// <summary>
/// Checks the share of wins and its Wilson score interval, against values computed by hand.
/// </summary>
[TestFixture]
public sealed class ModelComparisonWinShareTests
{
private const double TOLERANCE = 0.001;
[Test]
public void WithoutDecidedRunsNothingIsClear()
{
var winShare = ModelComparisonWinShare.Of(0, 0);
using (Assert.EnterMultipleScope())
{
Assert.That(winShare.Share, Is.Null);
Assert.That(winShare.LowerBound, Is.Null);
Assert.That(winShare.UpperBound, Is.Null);
Assert.That(winShare.Significance, Is.EqualTo(ModelComparisonSignificance.NOT_YET_CLEAR));
}
}
[Test]
public void EightOfTenIsNotYetClear()
{
var winShare = ModelComparisonWinShare.Of(8, 2);
using (Assert.EnterMultipleScope())
{
Assert.That(winShare.Share, Is.EqualTo(0.8).Within(TOLERANCE));
Assert.That(winShare.LowerBound, Is.EqualTo(0.4902).Within(TOLERANCE));
Assert.That(winShare.UpperBound, Is.EqualTo(0.9433).Within(TOLERANCE));
Assert.That(winShare.Significance, Is.EqualTo(ModelComparisonSignificance.NOT_YET_CLEAR), "The interval still reaches below one half.");
}
}
[Test]
public void EighteenOfTwentyIsClearlyBetter()
{
var winShare = ModelComparisonWinShare.Of(18, 2);
using (Assert.EnterMultipleScope())
{
Assert.That(winShare.LowerBound, Is.EqualTo(0.6990).Within(TOLERANCE));
Assert.That(winShare.Significance, Is.EqualTo(ModelComparisonSignificance.CLEARLY_BETTER));
}
}
[Test]
public void NoneOfTenIsClearlyWorse()
{
var winShare = ModelComparisonWinShare.Of(0, 10);
using (Assert.EnterMultipleScope())
{
Assert.That(winShare.LowerBound, Is.EqualTo(0).Within(TOLERANCE));
Assert.That(winShare.UpperBound, Is.EqualTo(0.2775).Within(TOLERANCE));
Assert.That(winShare.Significance, Is.EqualTo(ModelComparisonSignificance.CLEARLY_WORSE));
}
}
[Test]
public void ANegativeCountIsRefused()
{
Assert.That(() => ModelComparisonWinShare.Of(-1, 3), Throws.TypeOf<ArgumentOutOfRangeException>());
}
}