public static void Main(string[] args)
{
WeightsReader wr = new WeightsReader("CifarWeight.csv", "CifarBias.csv");
Console.WriteLine("Generating encryption keys {0}", DateTime.Now);
var factory = new EncryptedSealBfvFactory(new ulong[] { 2148728833, 2148794369, 2149810177 }, 16384, DecompositionBitCount: 60, GaloisDecompositionBitCount: 60, SmallModulusCount: 7);
Console.WriteLine("Encryption keys ready {0}", DateTime.Now);
string fileName = "cifar-test.tsv";
var readerLayer = new LLConvReader
{
FileName = fileName,
SparseFormat = false,
InputShape = new int[] { 3, 32, 32 },
KernelShape = new int[] { 3, 8, 8 },
Upperpadding = new int[] { 0, 1, 1 },
Lowerpadding = new int[] { 0, 1, 1 },
Stride = new int[] { 1000, 2, 2 },
NormalizationFactor = 1.0,
Scale = 128.0
};
var encryptLayer = new EncryptLayer()
{
Source = readerLayer, Factory = factory
};
var convLayer1 = new LLPoolLayer()
{
Source = encryptLayer,
InputShape = new int[] { 3, 32, 32 },
KernelShape = new int[] { 3, 8, 8 },
Upperpadding = new int[] { 0, 1, 1 },
Lowerpadding = new int[] { 0, 1, 1 },
Stride = new int[] { 1000, 2, 2 },
MapCount = new int[] { 83, 1, 1 },
WeightsScale = 256.0,
Weights = (double[])wr.Weights[0],
Bias = (double[])wr.Biases[0]
};
var VectorizeLayer2 = new LLVectorizeLayer()
{
Source = convLayer1
};
var activationLayer3 = new SquareActivation()
{
Source = VectorizeLayer2
};
var convEngine = new ConvolutionEngine()
{
InputShape = new int[] { 83, 14, 14 },
KernelShape = new int[] { 83, 6, 6 },
Padding = new bool[] { false, false, false },
Stride = new int[] { 83, 2, 2 },
MapCount = new int[] { 163, 1, 1 }
};
var denseLayer4 = new LLDenseLayer
{
Source = activationLayer3,
WeightsScale = 512.0,
Weights = convEngine.GetDenseWeights((double[])wr.Weights[1]),
Bias = convEngine.GetDenseBias((double[])wr.Biases[1]),
InputFormat = EVectorFormat.dense,
ForceDenseFormat = true
};
var activationLayer5 = new SquareActivation()
{
Source = denseLayer4
};
var denseLayer6 = new LLDenseLayer()
{
Source = activationLayer5,
Weights = (double[])wr.Weights[2],
Bias = (double[])wr.Biases[2],
WeightsScale = 1024.0,
InputFormat = EVectorFormat.dense
};
var network = denseLayer6;
Console.WriteLine("Preparing");
// Visualize layer construction
for (var p = (INetwork)network; p != null; p = p.Source)
{
if (p is BaseLayer b)
{
b.Verbose = true;
}
}
network.PrepareNetwork();
var m = network.GetNext();
Utils.Show(m, factory, readerLayer.Labels);
Console.WriteLine("Max computed value {0} ({1})", RawMatrix.Max, Math.Log(RawMatrix.Max) / Math.Log(2));
}
public static Tuple<INetwork, IFactory> LargeLoLa(string FileName, bool Encrypt)
{
Console.WriteLine("Large LoLa mode");
WeightsReader wr = new WeightsReader("MnistLargeWeight.csv", "MnistLargeBias.csv");
Console.Write("Generating keys in ");
var start = DateTime.Now;
var Factory = Encrypt ? (IFactory)new EncryptedSealBfvFactory(new ulong[] { 2148728833, 2148794369, 2149810177 }, 16384, DecompositionBitCount: 60, GaloisDecompositionBitCount: 60, SmallModulusCount: 7)
: new RawFactory(16384);
var end = DateTime.Now;
Console.WriteLine("{0} seconds", (end - start).TotalSeconds);
var readerLayer = new LLConvReader
{
FileName = FileName,
SparseFormat = true,
InputShape = new int[] { 1, 28, 28 },
KernelShape = new int[] { 1, 8, 8 },
Upperpadding = new int[] { 0, 1, 1 },
Lowerpadding = new int[] { 0, 1, 1 },
Stride = new int[] { 1000, 2, 2 },
NormalizationFactor = 1.0,
Scale = 16.0
};
var encryptLayer = new EncryptLayer() { Source = readerLayer, Factory = Factory };
var convLayer1 = new LLPoolLayer()
{
Source = encryptLayer,
InputShape = new int[] { 1, 28, 28 },
KernelShape = new int[] { 1, 8, 8 },
Upperpadding = new int[] { 0, 1, 1 },
Lowerpadding = new int[] { 0, 1, 1 },
Stride = new int[] { 1000, 2, 2 },
MapCount = new int[] { 83, 1, 1 },
WeightsScale = 4096,
Weights = ((double[])wr.Weights[0]).Select(x => x/256).ToArray(),
Bias = (double[])wr.Biases[0]
};
var VectorizeLayer2 = new LLVectorizeLayer() { Source = convLayer1 };
//var activationLayer3 = new SquareActivation() { Source = VectorizeLayer2 };
var activationLayer3 = new AppxReLUActivation() { Source = VectorizeLayer2 };
//var activationLayer3 = new LeakyReLUActivation() { Source = VectorizeLayer2 };
//var activationLayer3 = new ReLUActivation() { Source = VectorizeLayer2 };
var convEngine = new ConvolutionEngine()
{
InputShape = new int[] { 83, 12, 12 },
KernelShape = new int[] { 83, 6, 6 },
Padding = new bool[] { false, false, false },
Stride = new int[] { 83, 2, 2 },
MapCount = new int[] { 163, 1, 1 }
};
var denseLayer4 = new LLDenseLayer
{
Source = activationLayer3,
WeightsScale = 64,
Weights = convEngine.GetDenseWeights((double[])wr.Weights[1]),
Bias = convEngine.GetDenseBias((double[])wr.Biases[1]),
InputFormat = EVectorFormat.dense,
ForceDenseFormat = true
};
//var activationLayer5 = new SquareActivation() { Source = denseLayer4 };
var activationLayer5 = new AppxReLUActivation() { Source = denseLayer4 };
//var activationLayer5 = new LeakyReLUActivation() { Source = denseLayer4 };
//var activationLayer5 = new ReLUActivation() { Source = denseLayer4 };
var denseLayer6 = new LLDenseLayer()
{
Source = activationLayer5,
Weights = (double[])wr.Weights[2],
Bias = (double[])wr.Biases[2],
WeightsScale = 512,
InputFormat = EVectorFormat.dense
};
return new Tuple<INetwork, IFactory>(denseLayer6, Factory);
}
public static void Main(string[] args)
{
var options = new Options();
var parsed = Parser.Default.ParseArguments <Options>(args).WithParsed(x => options = x);
if (parsed.Tag == ParserResultType.NotParsed)
{
Environment.Exit(-2);
}
WeightsReader wr = new WeightsReader("CifarWeight.csv", "CifarBias.csv");
// Model has accuracy of 76.5
// Current parameters (scale) provide accuracy of 76.31% and uses 78.55 + 1 bits in message length
// It has a latency of 740 seconds on the reference machine (Azure B8ms server at rest)
Console.WriteLine("Generating encryption keys {0}", DateTime.Now);
IFactory factory = null;
if (options.Encrypt)
{
factory = new EncryptedSealBfvFactory(new ulong[] { 957181001729, 957181034497 }, 16384, DecompositionBitCount: 60, GaloisDecompositionBitCount: 60, SmallModulusCount: 8);
}
else
{
factory = new RawFactory(16 * 1024);
}
Console.WriteLine("Encryption keys ready {0}", DateTime.Now);
int numberOfRecords = 10000;
bool verbose = options.Verbose;
string fileName = "cifar-test.tsv";
var readerLayer = new LLConvReader
{
FileName = fileName,
SparseFormat = false,
InputShape = new int[] { 3, 32, 32 },
KernelShape = new int[] { 3, 8, 8 },
Upperpadding = new int[] { 0, 1, 1 },
Lowerpadding = new int[] { 0, 1, 1 },
Stride = new int[] { 1000, 2, 2 },
NormalizationFactor = 1.0 / 256.0,
Scale = 8,
Verbose = verbose
};
var EncryptLayer = new EncryptLayer()
{
Source = readerLayer, Factory = factory
};
var StartTimingLayer = new TimingLayer()
{
Source = EncryptLayer, StartCounters = new string[] { "Inference-Time" }
};
var ConvLayer1 = new LLPoolLayer()
{
Source = StartTimingLayer,
InputShape = new int[] { 3, 32, 32 },
KernelShape = new int[] { 3, 8, 8 },
Upperpadding = new int[] { 0, 1, 1 },
Lowerpadding = new int[] { 0, 1, 1 },
Stride = new int[] { 1000, 2, 2 },
MapCount = new int[] { 83, 1, 1 },
WeightsScale = 256.0,
Weights = (double[])wr.Weights[0],
Bias = (double[])wr.Biases[0],
Verbose = verbose
};
var VectorizeLayer2 = new LLVectorizeLayer()
{
Source = ConvLayer1,
Verbose = verbose
};
var ActivationLayer3 = new SquareActivation()
{
Source = VectorizeLayer2,
Verbose = verbose
};
var ConvEngine = new ConvolutionEngine()
{
InputShape = new int[] { 83, 14, 14 },
KernelShape = new int[] { 83, 10, 10 },
Upperpadding = new int[] { 0, 4, 4 },
Lowerpadding = new int[] { 0, 4, 4 },
Stride = new int[] { 83, 2, 2 },
MapCount = new int[] { 112, 1, 1 }
};
var DenseLayer4 = new LLDenseLayer
{
Source = ActivationLayer3,
WeightsScale = 512.0,
Weights = ConvEngine.GetDenseWeights((double[])wr.Weights[1]),
Bias = ConvEngine.GetDenseBias((double[])wr.Biases[1]),
InputFormat = EVectorFormat.dense,
ForceDenseFormat = true,
Verbose = verbose
};
var ActivationLayer5 = new SquareActivation()
{
Source = DenseLayer4,
Verbose = verbose
};
var DenseLayer6 = new LLDenseLayer()
{
Source = ActivationLayer5,
Weights = (double[])wr.Weights[2],
Bias = (double[])wr.Biases[2],
WeightsScale = 512.0,
InputFormat = EVectorFormat.dense,
Verbose = verbose
};
var StopTimingLayer = new TimingLayer()
{
Source = DenseLayer6, StopCounters = new string[] { "Inference-Time" }
};
var network = StopTimingLayer;
Console.WriteLine("Preparing");
network.PrepareNetwork();
int count = 0;
int errs = 0;
int batchSize = 1;
while (count < numberOfRecords)
{
using (var m = network.GetNext())
{
Utils.ProcessInEnv(env =>
{
var decrypted = m.Decrypt(env);
int pred = 0;
for (int j = 1; j < decrypted.RowCount; j++)
{
if (decrypted[j, 0] > decrypted[pred, 0])
{
pred = j;
}
}
if (pred != readerLayer.Labels[0])
{
errs++;
}
count++;
if (count % batchSize == 0)
{
Console.Write("errs {0}/{1} accuracy {2:0.000}% prediction {3} label {4} {5}ms", errs, count, 100 - (100.0 * errs / (count)), pred, readerLayer.Labels[0], TimingLayer.GetStats());
if (options.Encrypt)
{
Console.WriteLine();
}
else
{
Console.WriteLine(" 2^{0} largest-value", Math.Log(RawMatrix.Max) / Math.Log(2));
}
}
}, factory);
}
}
Console.WriteLine("errs {0}/{1} accuracy {2:0.000}%", errs, count, 100 - (100.0 * errs / (count)));
network.DisposeNetwork();
if (!options.Encrypt)
{
Console.WriteLine("Max computed value 2^{1}", Math.Log(RawMatrix.Max) / Math.Log(2));
}
}
