public static TestCaseParameters GetTestOption(TestCaseOption testCaseOption, AlgoUsed algoUsed)
{
TestCaseParameters testCaseParameters = null;
List <double> xArray;
List <double> yArray;
int defaultValue = -1;
if (algoUsed == AlgoUsed.RBF)
{
defaultValue = 0;
}
switch (testCaseOption)
{
case TestCaseOption.LINEAR_SIMPLE:
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
2, 1
},
nplSize = 2,
X = new List <double> {
1, 1, 2, 3, 3, 3
},
Y = new List <double> {
defaultValue, 1, 1
},
sampleSize = 3,
isClassification = true,
nbCentroids = 2,
nbClasses = 2,
maxKmeans = 100,
gamma = 0.1f
});
case TestCaseOption.LINEAR_MULTIPLE:
xArray = new List <double>();
yArray = new List <double>();
for (int i = 0; i < 100; ++i)
{
xArray.Add(Random.Range(0.0f, 1.0f) + (i < 50 ? 1 : 2));
xArray.Add(Random.Range(0.0f, 1.0f) + (i < 50 ? 1 : 2));
yArray.Add(i < 50 ? 1 : 0);
}
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
2, 1
},
nplSize = 2,
X = xArray,
Y = yArray,
sampleSize = 100,
isClassification = true,
nbCentroids = 2,
nbClasses = 2,
maxKmeans = 1000,
gamma = 0.5f
});
case TestCaseOption.XOR:
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
2, 3, 1
},
nplSize = 3,
X = new List <double> {
0, 0, 0, 1, 1, 0, 1, 1
},
Y = new List <double> {
1, defaultValue, defaultValue, 1
},
sampleSize = 4,
isClassification = true,
nbCentroids = 4,
nbClasses = 2,
maxKmeans = 100,
gamma = 1f
});
case TestCaseOption.CROSS:
xArray = new List <double>();
yArray = new List <double>();
for (int i = 0; i < 500; ++i)
{
xArray.Add(Random.Range(-1.0f, 1.0f));
xArray.Add(Random.Range(-1.0f, 1.0f));
yArray.Add(Mathf.Abs((float)xArray[i * 2]) <= 0.3 || Mathf.Abs((float)xArray[i * 2 + 1]) <= 0.3 ? 1 : defaultValue);
}
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
2, 4, 1
},
nplSize = 3,
X = xArray,
Y = yArray,
sampleSize = 500,
isClassification = true,
nbCentroids = 9,
nbClasses = 2,
maxKmeans = 1000,
gamma = 0.01f
});
case TestCaseOption.MULTI_LINEAR_3_CLASSES:
xArray = new List <double>();
yArray = new List <double>();
for (int i = 0; i < 500; ++i)
{
xArray.Add(Random.Range(0f, 1.0f));
xArray.Add(Random.Range(0f, 1.0f));
int xIndex = i * 2;
int yIndex = i * 2 + 1;
if (-xArray[xIndex] - xArray[yIndex] - 0.5 > 0 && xArray[yIndex] < 0 &&
xArray[xIndex] - xArray[yIndex] - 0.5 < 0)
{
yArray.Add(1);
yArray.Add(defaultValue);
yArray.Add(defaultValue);
}
else if (-xArray[xIndex] - xArray[yIndex] - 0.5 < 0 && xArray[yIndex] > 0 &&
xArray[xIndex] - xArray[yIndex] - 0.5 < 0)
{
yArray.Add(defaultValue);
yArray.Add(1);
yArray.Add(defaultValue);
}
else if (-xArray[xIndex] - xArray[yIndex] - 0.5 < 0 && xArray[yIndex] < 0 &&
xArray[xIndex] - xArray[yIndex] - 0.5 > 0)
{
yArray.Add(defaultValue);
yArray.Add(defaultValue);
yArray.Add(1);
}
else
{
yArray.Add(defaultValue);
yArray.Add(defaultValue);
yArray.Add(defaultValue);
}
}
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
2, 3
},
nplSize = 2,
X = xArray,
Y = yArray,
sampleSize = 500,
isClassification = true,
nbCentroids = 4,
nbClasses = 2,
maxKmeans = 100,
gamma = 0.1f
});
case TestCaseOption.MULTI_CROSS:
xArray = new List <double>();
yArray = new List <double>();
for (int i = 0; i < 1000; ++i)
{
xArray.Add(Random.Range(-1.0f, 1.0f));
xArray.Add(Random.Range(-1.0f, 1.0f));
int xIndex = i * 2;
int yIndex = i * 2 + 1;
if (Math.Abs(xArray[xIndex]) % 0.5 <= 0.25 && Math.Abs(xArray[yIndex]) % 0.5 > 0.25)
{
yArray.Add(1);
yArray.Add(defaultValue);
yArray.Add(defaultValue);
}
else if (Math.Abs(xArray[xIndex]) % 0.5 > 0.25 && Math.Abs(xArray[yIndex]) % 0.5 <= 0.25)
{
yArray.Add(defaultValue);
yArray.Add(1);
yArray.Add(defaultValue);
}
else
{
yArray.Add(defaultValue);
yArray.Add(defaultValue);
yArray.Add(1);
}
}
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
2, 3, 3, 3
},
nplSize = 4,
X = xArray,
Y = yArray,
sampleSize = 1000,
isClassification = true,
nbCentroids = 5,
nbClasses = 2,
maxKmeans = 100,
gamma = 0.1f
});
case TestCaseOption.LINEAR_SIMPLE_2D:
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
1, 1
},
nplSize = 2,
X = new List <double> {
1, 2
},
Y = new List <double> {
2, 3
},
sampleSize = 2,
isClassification = false,
nbCentroids = 2,
nbClasses = 4,
maxKmeans = 100,
gamma = 0.1f
});
case TestCaseOption.NON_LINEAR_SIMPLE_2D:
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
1, 3, 1
},
nplSize = 3,
X = new List <double> {
1, 2, 3
},
Y = new List <double> {
2, 4, 3
},
sampleSize = 3,
isClassification = false,
nbCentroids = 5,
nbClasses = 5,
maxKmeans = 100,
gamma = 0.1f
});
case TestCaseOption.LINEAR_SIMPLE_3D:
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
2, 1
},
nplSize = 2,
X = new List <double> {
1, 1, 2, 2, 3, 1
},
Y = new List <double> {
2, 4, 3
},
sampleSize = 3,
isClassification = false,
nbCentroids = 2,
nbClasses = 5,
maxKmeans = 100,
gamma = 0.1f
});
case TestCaseOption.LINEAR_TRICKY_3D:
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
2, 1
},
nplSize = 2,
X = new List <double> {
1, 1, 2, 2, 3, 3
},
Y = new List <double> {
1, 2, 3
},
sampleSize = 3,
isClassification = false,
nbCentroids = 2,
nbClasses = 4,
maxKmeans = 100,
gamma = 0.1f
});
case TestCaseOption.NON_LINEAR_SIMPLE_3D:
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
2, 2, 1
},
nplSize = 3,
X = new List <double> {
1, 0, 0, 1, 1, 1, 0, 0
},
Y = new List <double> {
2, 1, 4, 3
},
sampleSize = 4,
isClassification = false,
nbCentroids = 4,
nbClasses = 5,
maxKmeans = 100,
gamma = 0.1f
});
case TestCaseOption.IMAGES_TEST:
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
2, 3, 1
},
nplSize = 3,
X = new List <double> {
255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255
},
Y = new List <double> {
defaultValue,
1,
1,
1,
defaultValue
},
sampleSize = 5,
isClassification = true,
nbCentroids = 2,
nbClasses = 2,
maxKmeans = 1000,
gamma = 0.1f
});
case TestCaseOption.USE_IMAGES:
return(new TestCaseParameters
{
neuronsPerLayer = new List <int> {
2, 5, 1
},
nplSize = 3,
sampleSize = 5,
isClassification = true,
nbCentroids = 2,
nbClasses = 2,
maxKmeans = 1000,
gamma = 100000f
});
}
return(testCaseParameters);
}
public static double[] RunMachineLearningTestCase(bool useLinearModel, TestCaseParameters param, int epochs,
double learningRate, TestCaseParameters simulateTestCaseParameters, AlgoUsed algoUsed,
List <double> outputToTest = null, int outputSize = 0)
{
lastTestCaseParameters = simulateTestCaseParameters ?? param;
if (simulateTestCaseParameters == null)
{
Debug.Log("Generate new test case");
}
if (lastTestCaseParameters == null)
{
return(null);
}
if (param.neuronsPerLayer[0] < 3)
{
mlcInstance.InstantiateSpheresInScene(lastTestCaseParameters.X, lastTestCaseParameters.sampleSize, null);
}
List <double> outputTest = outputToTest == null ? param.X : outputToTest;
int outputTotalSize = outputToTest == null ? lastTestCaseParameters.sampleSize * lastTestCaseParameters.neuronsPerLayer[lastTestCaseParameters.nplSize - 1] : outputSize;
double[] result = new double[outputTotalSize];
IntPtr rawResut;
if (useLinearModel)
{
rawResut = CppImporter.trainLinearModel(
lastTestCaseParameters.X.ToArray(),
lastTestCaseParameters.neuronsPerLayer[0],
lastTestCaseParameters.Y.ToArray(),
lastTestCaseParameters.neuronsPerLayer[lastTestCaseParameters.nplSize - 1],
lastTestCaseParameters.sampleSize,
epochs,
learningRate,
lastTestCaseParameters.isClassification
);
}
else
{
rawResut = CppImporter.trainMLPModel(
lastTestCaseParameters.neuronsPerLayer.ToArray(),
lastTestCaseParameters.nplSize,
lastTestCaseParameters.X.ToArray(),
lastTestCaseParameters.Y.ToArray(),
lastTestCaseParameters.sampleSize,
epochs,
learningRate,
lastTestCaseParameters.isClassification,
outputTest.ToArray(),
outputTotalSize
);
}
Marshal.Copy(rawResut, result, 0, outputTotalSize);
Debug.Log("Result MLP or Linear");
return(result);
}
