Save and Load Serialization. about convnetsharp

[cbovar]

I will look at that when I'll be back from holidays

[cbovar]

Serialization tests should be made generic to detect such error earlier

[Diapa]

I have this problem. How do I fix it? I installed with nuget and I use regressionlayer. The exception seems to come from

ConvNetSharp/src/ConvNetSharp.Core/Layers/LayerBase.cs

Line 101 in f89f964

var t = Activator.CreateInstance(type, dico) as LayerBase<T>;

[cbovar]

Hi @Diapa ,
Could you try with the new nuget package (0.4.3) ?

[Diapa]

Yes it works now thank you, but using this version GPU doesn't work. I get the same problem, but only using GPU. Additionally I get an ArgumentNullException when adding a convolutional layer to the network (only GPU version). It comes from ConvNetSharp.Volume.GPU.Single.Volume.LoadKernels().

[cbovar]

Do you have a minimal code that reproduces this error?

[Diapa]

Maybe there's some problem with my CUDA/cudnn installation if you don't get this problem... But I was able to run GPU version with v0.4.2.

using ConvNetSharp.Core;
using ConvNetSharp.Core.Layers;
using ConvNetSharp.Volume.GPU.Single;

namespace m
{
    class Program
    {
        static void Main(string[] args)
        {
            BuilderInstance.Volume = new VolumeBuilder();

            Net<float> net = new Net<float>();
            net.AddLayer(new InputLayer<float>(20, 20, 3));
            net.AddLayer(new ConvLayer<float>(5, 5, 48)); //System.ArgumentNullException
            /*
            System.IO.StreamReader..ctor(Stream stream, Encoding encoding, Boolean detectEncodingFromByteOrderMarks, Int32 bufferSize, Boolean leaveOpen)
            System.IO.StreamReader..ctor(Stream stream)
            ConvNetSharp.Volume.GPU.KernelLoader`1.LoadKernel(String name, Stream stream)
            ConvNetSharp.Volume.GPU.Single.Volume.LoadKernels()
            ConvNetSharp.Volume.GPU.Single.VolumeBuilder.Random(Shape shape, Double mu, Double std)
            ConvNetSharp.Core.Layers.ConvLayer`1.UpdateOutputSize()
            ConvNetSharp.Core.Net`1.AddLayer(LayerBase`1 layer)
            [row 15 in this file]
            */
        }
    }
}