VS 2017 |
VS 2015 |
Nuget |
---|---|---|
Started initially as C# port of ConvNetJS. You can use ConvNetSharp to train and evaluate convolutional neural networks (CNN).
Thank you very much to the original author of ConvNetJS (Andrej Karpathy) and to all the contributors!
🔹 NEW 🔹 See ConvNetSharp in action driving a car ! Micromachine.AI
Core.Layers | Flow.Layers | Computation graph |
---|---|---|
No computation graph | Layers that create a computation graph behind the scene | 'Pure flow' |
Network organised by stacking layers | Network organised by stacking layers | 'Ops' connected to each others. Can implement more complex networks |
![]() |
![]() |
![]() |
E.g. MnistDemo | E.g. MnistFlowGPUDemo or Flow version of Classify2DDemo | E.g. ExampleCpuSingle |
Here's a minimum example of defining a 2-layer neural network and training it on a single data point:
using System;
using ConvNetSharp.Core;
using ConvNetSharp.Core.Layers.Double;
using ConvNetSharp.Core.Training.Double;
using ConvNetSharp.Volume;
using ConvNetSharp.Volume.Double;
namespace MinimalExample
{
internal class Program
{
private static void Main()
{
// specifies a 2-layer neural network with one hidden layer of 20 neurons
var net = new Net<double>();
// input layer declares size of input. here: 2-D data
// ConvNetJS works on 3-Dimensional volumes (width, height, depth), but if you're not dealing with images
// then the first two dimensions (width, height) will always be kept at size 1
net.AddLayer(new InputLayer(1, 1, 2));
// declare 20 neurons
net.AddLayer(new FullyConnLayer(20));
// declare a ReLU (rectified linear unit non-linearity)
net.AddLayer(new ReluLayer());
// declare a fully connected layer that will be used by the softmax layer
net.AddLayer(new FullyConnLayer(10));
// declare the linear classifier on top of the previous hidden layer
net.AddLayer(new SoftmaxLayer(10));
// forward a random data point through the network
var x = BuilderInstance.Volume.From(new[] { 0.3, -0.5 }, new Shape(2));
var prob = net.Forward(x);
// prob is a Volume. Volumes have a property Weights that stores the raw data, and WeightGradients that stores gradients
Console.WriteLine("probability that x is class 0: " + prob.Get(0)); // prints e.g. 0.50101
var trainer = new SgdTrainer(net) { LearningRate = 0.01, L2Decay = 0.001 };
trainer.Train(x, BuilderInstance.Volume.From(new[] { 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 }, new Shape(1, 1, 10, 1))); // train the network, specifying that x is class zero
var prob2 = net.Forward(x);
Console.WriteLine("probability that x is class 0: " + prob2.Get(0));
// now prints 0.50374, slightly higher than previous 0.50101: the networks
// weights have been adjusted by the Trainer to give a higher probability to
// the class we trained the network with (zero)
}
}
}
Fluent API (see FluentMnistDemo)
var net = FluentNet<double>.Create(24, 24, 1)
.Conv(5, 5, 8).Stride(1).Pad(2)
.Relu()
.Pool(2, 2).Stride(2)
.Conv(5, 5, 16).Stride(1).Pad(2)
.Relu()
.Pool(3, 3).Stride(3)
.FullyConn(10)
.Softmax(10)
.Build();
To switch to GPU mode:
- add '
GPU
' in the namespace:using ConvNetSharp.Volume.
GPU.Single;
orusing ConvNetSharp.Volume.
GPU.Double;
- add
BuilderInstance<float>.Volume = new ConvNetSharp.Volume.GPU.Single.VolumeBuilder();
orBuilderInstance<double>.Volume = new ConvNetSharp.Volume.GPU.Double.VolumeBuilder();
at the beggining of your code
You must have CUDA version 8 and Cudnn version 6.1 installed. Cudnn bin path should be referenced in the PATH environment variable.
Mnist GPU demo here
// Serialize to json
var json = net.ToJsonN();
// Deserialize from json
Net deserialized = SerializationExtensions.FromJson<double>(json);