LayerCloning
Layers · AiDotNet.NeuralNetworks.Layers
Cloning for layers, built on the construction state layers already record for serialization.
For Beginners
layer.Clone() gives you a separate copy of a layer, including what
it has learned. You do not write anything to make this work on a layer of your own: mark the
constructor arguments the layer needs with [LayerState] — which it already needs for
saving and loading — and cloning follows.
How It Works
This deliberately reuses the generated construction manifest and
GeneratedLayerFactories.TryCreate rather than introducing a second reconstruction
mechanism. Durable persistence and the in-memory object channel are emitted from the same
[LayerState] model; the live channel avoids formatting constructor-owned child-layer
weights during a clone. Both call the layer’s real constructor with the values it was originally
given, and the build already fails for a layer whose required state cannot be sourced. Sharing
the generated model means saving, loading, and cloning cannot silently drift into independent
constructor recipes.
The learned parameters travel separately, through GetParameters and
UpdateParameters. That is the contract training exercises on every step, so a clone
cannot disagree with training about what the parameters are.
Methods
| Method | Summary |
|---|---|
ILayer<T> Clone(LayerBase<T> source, CloneOptions options) |
Creates an independent copy of a layer. |