public void PredictWinner_PositiveWins()
{
// Given
var bags = new[]
{
new BagTestData(1, 2, new[]
{
new[] {26.4386, 25.5379}
}),
new BagTestData(1, 3, new[]
{
new[] {44.9671, 32.1093}
}),
new BagTestData(-1, 4, new[]
{
new[] {20.7886, 17.2379}
})
};
var bagToTest = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}},
new Dictionary<int, float> {{1, 59.1986f}, {2, 34.5393f}}
});
int predictedLabelForOne = PredictBagLabel(bags, bagToTest, 2);
// Then
Assert.That(predictedLabelForOne, Is.EqualTo(+1));
}
public void PredictWinner_SimpleRuleSet()
{
// Given
var bagsToTrain = new SparseBag[0];
var bagsToTest = new[]
{
new SparseBag(-1, "1", new[]
{
new Dictionary<int, float> {{1, 1.0f}, {2, 1.9f}}
}),
new SparseBag(+1, "2", new[]
{
new Dictionary<int, float> {{1, 1.0f}, {2, 3.0f}}
})
};
var classifier = ClassifierFactory.GetMultiInstanceClassifier("rules:class+1=1>0.9^2>2.0");
// When
classifier.Train(bagsToTrain);
foreach (var bag in bagsToTest)
{
var predictedClass = classifier.PredictWinner(bag);
// Then
Assert.That(predictedClass, Is.EqualTo(bag.Label));
}
}
public void PredictWinner_SparseBagsAreSingletons_NegativeWins()
{
// Given
var bags = new[]
{
new BagTestData(-1, 2, new[]
{
new[] {26.4386, 25.5379}
}),
new BagTestData(-1, 3, new[]
{
new[] {44.9671, 32.1093}
}),
new BagTestData(+1, 4, new[]
{
new[] {20.7886, 17.2379}
})
};
var bagToTest = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}},
new Dictionary<int, float> {{1, 59.1986f}, {2, 34.5393f}}
});
var miClassifier = ClassifierFactory.GetMultiInstanceClassifier("mi:model=svm,gamma=123");
// When
var predictedLabel = PredictBagLabel(miClassifier, bags, bagToTest);
// Then
Assert.That(predictedLabel, Is.EqualTo(-1));
}
public void Train_Test_NegativeWins()
{
// Given
var bags = new[]
{
new BagTestData(+1, 1, new[]
{
new[] {22.8529, 22.6664},
new[]{59.1986, 34.5393}
}),
new BagTestData(-1, 2, new[]
{
new[] {26.4386, 25.5379}
}),
new BagTestData(-1, 3, new[]
{
new[] {44.9671, 32.1093}
}),
new BagTestData(1, 4, new[]
{
new[] {20.7886, 17.2379}
})
};
var bagToTest = new SparseBag(+1, "1", new[]
{
new Dictionary<int, float> {{1, 44.9671f}, {2, 32.1093f}}
});
var predictedLabelForOne = PredictBagLabel(bags, bagToTest, 1);
// Then
Assert.That(predictedLabelForOne, Is.EqualTo(-1));
}
public void Write_Verify()
{
var tmpFileName = Path.GetTempFileName();
try
{
// Given
var bag1 = new SparseBag(1, "bag1", new[] {new Dictionary<int, float> {{1, 0.5f}}});
var bag2 = new SparseBag(1, "bag2", new[] { new Dictionary<int, float> { { 1, 0.5f } }, new Dictionary<int, float> { { 2, 0.6f }, { 3, 1.6f } } });
var sparseBagWriter = DatasetWriterFactory.GetDatasetWriter<SparseBag>(tmpFileName);
// When
sparseBagWriter.Write(bag1);
sparseBagWriter.Write(bag2);
sparseBagWriter.Dispose();
var sparseBagReader = DatasetReaderFactory.GetDatasetReader<SparseBag>(tmpFileName);
var readBag1 = sparseBagReader.Next();
var readBag2 = sparseBagReader.Next();
var readBag3 = sparseBagReader.Next();
// Then
Assert.That(readBag1, Is.EqualTo(bag1));
Assert.That(readBag2, Is.EqualTo(bag2));
Assert.That(readBag3, Is.Null);
}
finally
{
if(File.Exists(tmpFileName))
{
File.Delete(tmpFileName);
}
}
}
private static int PredictBagLabel(IClassifier<SparseBag> classifier, BagTestData[] bags, SparseBag bagToTest)
{
var sparseBags = BagTestData.ToSparseBags(bags);
classifier.Train(sparseBags);
return classifier.PredictWinner(bagToTest);
}
public void HasValidIstances_True_OnlyNotNullFeatures()
{
// Given
var bag = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}}
});
// When/Then
Assert.That(bag.HasValidInstances, Is.True);
}
public void HasValidIstances_False()
{
// Given
var bag = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 0.0f}, {2, 0.0f}}
});
// When/Then
Assert.That(bag.HasValidInstances, Is.False);
}
public void Distance_Reflexive()
{
// Given
var bag1 = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}},
new Dictionary<int, float> {{1, 59.1986f}, {2, 34.5393f}}
});
// When
var distance = m_hausdorffDistance.Distance(bag1, bag1);
// Then
Assert.That(distance, Is.EqualTo(0.0f));
}
static internal IEnumerable<SparseBag> ToSparseBags(BagTestData[] bags)
{
var sparseBags = new SparseBag[bags.Length];
for (var bagId = 0; bagId < sparseBags.Length; bagId++)
{
var istances = new IDictionary<int, float>[bags[bagId].Points.Length];
for (var pointId = 0; pointId < bags[bagId].Points.Length; pointId++)
{
var j = 1;
var features = (from f in bags[bagId].Points[pointId] select f).ToDictionary(i => (j++), i => (float)i);
istances[pointId] = features;
}
sparseBags[bagId] = new SparseBag(bags[bagId].Label, bags[bagId].Name.ToString(), istances);
}
return sparseBags;
}
public void RemoveEmptyIstances_WithNothingToRemove()
{
// Given
var bag = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}},
});
var toObtain = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}}
});
// When
var simplifiedIstance = bag.RemoveEmptyIstances();
// Then
Assert.That(simplifiedIstance, Is.EqualTo(toObtain));
}
public void PredictWinner_EmptyRuleSet_PositiveClassWins()
{
// Given
var bagsToTrain = new SparseBag[0];
var bagsToTest = new SparseBag(+1, "2", new[]
{
new Dictionary<int, float> {{1, 1.0f}, {2, 3.0f}}
});
var classifier = ClassifierFactory.GetMultiInstanceClassifier("rules:class+1=");
// When
classifier.Train(bagsToTrain);
var predictedClass = classifier.PredictWinner(bagsToTest);
// Then
Assert.That(predictedClass, Is.EqualTo(bagsToTest.Label));
}
public void Distance_RealCase_1()
{
var istances1 = new[]
{
new Dictionary<int, float>
{
{1, 1}
},
new Dictionary<int, float>
{
{1, 2}
},
new Dictionary<int, float>
{
{1, 3}
}
};
var bag1 = new SparseBag(0, "a", istances1);
var istances2 = new[]
{
new Dictionary<int, float>
{
{1, 4}
},
new Dictionary<int, float>
{
{1, 5}
},
new Dictionary<int, float>
{
{1, 6}
}
};
var bag2 = new SparseBag(0, "a", istances2);
// When
var distance = m_minHausdorffDistance.Distance(bag1, bag2);
// Then
Assert.That(distance, Is.EqualTo(1));
}
public void Distance_Zero()
{
// Given
var bag1 = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 1.0f}, {2, 2.0f}},
new Dictionary<int, float> {{1, 100.0f}, {2, 200.0f}}
});
var bag2 = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 200.0f}, {2, 100.0f}},
new Dictionary<int, float> {{1, 1.0f}, {2, 2.0f}},
});
// When
var distanceA = m_hausdorffDistance.Distance(bag1, bag2);
var distanceB = m_hausdorffDistance.Distance(bag2, bag1);
// Then
Assert.That(distanceA, Is.EqualTo(distanceB));
}
public void Distance_Symmetric()
{
// Given
var bag1 = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}},
new Dictionary<int, float> {{1, 59.1986f}, {2, 34.5393f}}
});
var bag2 = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 0.0f}, {2, 22.6664f}},
new Dictionary<int, float> {{1, 59.1986f}, {2, 0.0f}}
});
// When
var distanceA = m_hausdorffDistance.Distance(bag1, bag2);
var distanceB = m_hausdorffDistance.Distance(bag2, bag1);
// Then
Assert.That(distanceA, Is.EqualTo(distanceB));
}
public void PredictWinner_NotBinary()
{
// Given
var bags = new[]
{
new BagTestData(-1, 2, new[]
{
new[] {26.4386, 25.5379}
}),
new BagTestData(1, 3, new[]
{
new[] {44.9671, 32.1093}
}),
new BagTestData(2, 4, new[]
{
new[] {20.7886, 17.2379}
})
};
var bagToTest = new SparseBag(-1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}},
new Dictionary<int, float> {{1, 59.1986f}, {2, 34.5393f}}
});
PredictBagLabel(bags, bagToTest, 3);
}
public void Equals_True()
{
// Given
var bag1 = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}},
new Dictionary<int, float> {{1, 59.1986f}, {2, 34.5393f}}
});
var bag2 = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}},
new Dictionary<int, float> {{1, 59.1986f}, {2, 34.5393f}}
});
// When / Then
Assert.That(bag1.Equals(bag2), Is.True);
}
private static int PredictBagLabel(BagTestData[] bags, SparseBag bagToTest, int K)
{
var sparseBags = BagTestData.ToSparseBags(bags);
// When
var citationNearestNeighbor = ClassifierFactory.GetMultiInstanceClassifier("nn:distance=hausdorff,K=" + K);
citationNearestNeighbor.Train(sparseBags);
return citationNearestNeighbor.PredictWinner(bagToTest);
}
public void EqualOperator_False()
{
// Given
var bag1 = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}},
new Dictionary<int, float> {{1, 59.1986f}, {2, 34.5393f}}
});
var bag2 = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}},
new Dictionary<int, float> {{1, 59.1386f}, {2, 34.5393f}}
});
// When / Then
Assert.That(bag1 != bag2);
}
public void Test_Musk1Speed()
{
// Given
const int K = 1;
var citationNearestNeighbor = ClassifierFactory.GetMultiInstanceClassifier("nn:distance=hausdorff,K=" + K);
const string datasetPath = "../../TestFiles/sparsebags/musk1.data";
var bagToTest = new SparseBag(-1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}},
new Dictionary<int, float> {{1, 59.1986f}, {2, 34.5393f}}
});
citationNearestNeighbor.Train(DatasetReaderFactory.GetDatasetReader<SparseBag>(datasetPath));
var start = DateTime.Now;
citationNearestNeighbor.PredictWinner(bagToTest);
Console.WriteLine("Total time: {0}", (DateTime.Now - start));
}
public void PredictWinner_NegativeWins_EmptyKnowledge()
{
// Given
var bags = new BagTestData[0];
var bagToTest = new SparseBag(1, "1", new[]
{
new Dictionary<int, float> {{1, 22.8529f}, {2, 22.6664f}},
new Dictionary<int, float> {{1, 59.1986f}, {2, 34.5393f}}
});
int predictedLabelForOne = PredictBagLabel(bags, bagToTest, 2);
// Then
Assert.That(predictedLabelForOne, Is.EqualTo(-1));
}
private static int PredictBagLabel(BagTestData[] bags, SparseBag bagToTest, int M)
{
var sparseBags = BagTestData.ToSparseBags(bags);
var citationNearestNeighbor = ClassifierFactory.GetMultiInstanceClassifier("rbfn:distance=hausdorff,M=" + M+",sigma=1");
citationNearestNeighbor.Train(sparseBags);
return citationNearestNeighbor.PredictWinner(bagToTest);
}