Esempio n. 1
0
        public void TSSetUp()
        {
            int[] layers = new int[] {
                1024, //32x32 input image
                4704, //6 @ 28x28 convolution
                1176, //6 @  14x14 subsampling
                1600, //16 @ 10x10 convolution
                400,  //16 @ 5x5 subsampling
                120,  // 120x1x1 convolution!
                84,   //full
                10    //full
            };
            LayerConnector[] map = new LayerConnector[] {
                new ConvolutionAuto(5, 6, 1, 4),
                new SubSampling(6),
                new Convolution(5, 16, 6, 4, GetSchema()),
                new SubSampling(16),
                new ConvolutionAuto(5, 120, 16, 0),
                new FullLayerConnector(),
                new FullLayerConnector()
            };
            ConvolutionalTopology  topology  = new ConvolutionalTopology(layers, 1, map);
            ConvolutionalGenerator generator = new ConvolutionalGenerator();

            Network = generator.Create(topology);
        }
Esempio n. 2
0
        void BuildNetwork()
        {
            int[] layers = new int[] {
                841, 1014, 1250, 100, 10
            };
            LayerConnector[] map = new LayerConnector[] {
                new ConvolutionAuto(5, 6, 1, 3),
                new ConvolutionAuto(5, 50, 6, 1),
                new FullLayerConnector(),
                new FullLayerConnector()
            };
            ConvolutionalTopology  topology  = new ConvolutionalTopology(layers, 1, map, new HyperbolicTangent());
            ConvolutionalGenerator generator = new ConvolutionalGenerator();

            Network             = generator.Create(topology);
            Network.LearnFactor = 0.0005;
            Network.Reset(-0.1, 0.1);
        }
Esempio n. 3
0
        public void ConvolutionSerialization()
        {
            int[] layers = new int[] {
                1024, //32x32 input image
                4704, //6 @ 28x28 convolution
                1176, //6 @  14x14 subsampling
                1600, //16 @ 10x10 convolution
                400,  //16 @ 5x5 subsampling
                120,  // 120x1x1 convolution!
                84,   //full
                10    //full
            };
            LayerConnector[] map = new LayerConnector[] {
                new ConvolutionAuto(5, 6, 1, 4),
                new SubSampling(6),
                new ConvolutionAuto(5, 16, 6, 4),
                new SubSampling(16),
                new ConvolutionAuto(5, 120, 16, 0),
                new FullLayerConnector(),
                new FullLayerConnector()
            };
            ConvolutionalTopology  topology  = new ConvolutionalTopology(layers, 1, map);
            ConvolutionalGenerator generator = new ConvolutionalGenerator();
            ConvolutionalNetwork   network   = generator.Create(topology);

            ConvolutionalXMLSerializer serializer = new ConvolutionalXMLSerializer();
            XDocument            doc      = serializer.Serialize(network);
            ConvolutionalNetwork network2 = serializer.Deserialize(doc);

            Assert.AreEqual(network2.Structure.Elements.Length, network.Structure.Elements.Length);
            Assert.AreEqual(network2.Structure.Elements[3000].GetDescription(), network.Structure.Elements[3000].GetDescription());
            Assert.AreEqual(network2.Structure.Elements[0].Next[0].Weight.Value, network.Structure.Elements[0].Next[0].Weight.Value);
            Assert.AreEqual(((NeuronBase)network2.Structure.Elements[4000]).Previous[0].Weight.Value, ((NeuronBase)network.Structure.Elements[4000]).Previous[0].Weight.Value);
            Assert.AreEqual(((NeuronBase)network2.Structure.Elements[3004]).Func.GetType(), ((NeuronBase)network.Structure.Elements[3004]).Func.GetType());

            Weight w1 = network2.Structure.Layers[0][0].Next[0].Weight;
            Weight w2 = network2.Structure.Layers[0][1].Next.Select(x => x.Weight).FirstOrDefault(x => x == w1);

            Assert.IsNotNull(w1);
            Assert.IsTrue(w1 == w2);

            Link[] l1 = network2.Structure.Layers[1];

            int c = GetConnectionsCount(l1, false, true);
            int w = GetWeightsCount(l1, false, true);

            Assert.AreEqual(c, 122304);
            Assert.AreEqual(w, 156);

            for (int i = 0; i < network.Structure.Layers.Length; i++)
            {
                for (int j = 0; j < network.Structure.Layers[i].Length; j++)
                {
                    Connection[] n1 = network.Structure.Layers[i][j].Next;
                    if (n1 != null)
                    {
                        Connection[] n2 = network2.Structure.Layers[i][j].Next;
                        Assert.IsNotNull(n2);
                        for (int z = 0; z < n1.Length; z++)
                        {
                            Assert.AreEqual(n1[z].Weight.Value, n2[z].Weight.Value);
                            Assert.AreEqual(n1[z].Weight.GetType(), n2[z].Weight.GetType());
                        }
                    }
                    if (network.Structure.Layers[i][j] is NeuronBase)
                    {
                        Connection[] p1 = ((NeuronBase)network.Structure.Layers[i][j]).Previous;
                        if (p1 != null)
                        {
                            Connection[] p2 = ((NeuronBase)network2.Structure.Layers[i][j]).Previous;
                            Assert.IsNotNull(p2);
                            for (int z = 0; z < p1.Length; z++)
                            {
                                Assert.AreEqual(p1[z].Weight.Value, p2[z].Weight.Value);
                                Assert.AreEqual(p1[z].Weight.GetType(), p2[z].Weight.GetType());
                            }
                        }
                    }
                }
            }
        }