public PoliciesPerformance Evaluate(uint learnedAction, int numActions, ContextualBanditLabel label)
{
return new PoliciesPerformance(
this.vw.Learn(
new LearnedVsConstantPolicy(learnedAction, numActions),
label,
VowpalWabbitPredictionType.Scalars));
}
public void TestNull3()
{
using (var vw = new VowpalWabbit<Context, ADF>("--cb_adf --rank_all --interact ac"))
{
var ctx = new Context()
{
ID = 25,
Vector = new float[] { 3 },
VectorC = new float[] { 2, 2, 3 },
ActionDependentFeatures = new[] {
new ADF {
ADFID = "23",
}
}.ToList()
};
var label = new ContextualBanditLabel() {
Action = 1,
Cost= 1,
Probability = 0.2f
};
vw.Learn(ctx, ctx.ActionDependentFeatures, 0, label);
ctx.Vector = null;
vw.Learn(ctx, ctx.ActionDependentFeatures, 0, label);
ctx.Vector = new float[] { 2 };
ctx.VectorC = null;
vw.Learn(ctx, ctx.ActionDependentFeatures, 0, label);
ctx.Vector = null;
vw.Learn(ctx, ctx.ActionDependentFeatures, 0, label);
}
}
public void TestNull5()
{
using (var vw = new VowpalWabbit<Context, ADF>("--cb_adf --rank_all --interact ab"))
{
var ctx = new Context()
{
ID = 25,
ActionDependentFeatures = new[] {
new ADF {
ADFID = "123"
},
new ADF(),
new ADF(),
new ADF { ADFID = "4"}
}.ToList()
};
var label = new ContextualBanditLabel()
{
Action = 1,
Cost = 1,
Probability = 0.2f
};
vw.Learn(ctx, ctx.ActionDependentFeatures, 0, label);
var result = vw.Predict(ctx, ctx.ActionDependentFeatures);
Assert.AreEqual(4, result.Length);
ctx.ActionDependentFeatures[0].ADFID = null;
ctx.ActionDependentFeatures[3].ADFID = null;
result = vw.Predict(ctx, ctx.ActionDependentFeatures);
Assert.AreEqual(4, result.Length);
}
}
private static EvalData Create(ContextualBanditLabel label, string policyName, uint actionTaken)
{
return new EvalData
{
PolicyName = policyName,
JSON = JsonConvert.SerializeObject(
new
{
name = policyName,
cost = VowpalWabbitContextualBanditUtil.GetUnbiasedCost(label.Action, actionTaken, label.Cost, label.Probability)
})
};
}