99. News
01/07/2017
- ConvNetSharp.Flow: A new way to create neural networks by defining a computation graph. There are now 3 ways of creating neural networks:
| Core.Layers | Flow.Layers | Pure Flow |
|---|---|---|
| No computation graph | Layers that create a computation graph behind the scene | Computation graph |
| 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 |
30/05/2017
- Available on Nuget in pre-release (i.e. not stable)
20/05/2017
- vs 2017 and vs 2015 solutions are now both on the same branch (using same source code).
27/03/2017
- Volumes have their own project
- Volumes have now 4 dimensions (width, height, channel, batchSize)
- Generic on numerics to use single or double precision (
Net<double>orNet<float>) - GPU implementation. Just add '
GPU' in the namespace:using ConvNetSharp.Volume.GPU.Single; - ConvNetSharp.Volume and ConvNetSharp.Core are on .NET Standard
- New way to serialize/deserialize. Basically Net object gives a nested dictionary that can be serialized the way you like.
Tag v0.2.0 was created just before commiting new version.