|
FlexNN 1
Fully connected neural network built from scratch with flexible n-layer design and multiple activations.
|
Per-activation hyperparameters for FlexNN::Activations. More...

Go to the source code of this file.
Classes | |
| struct | FlexNN::Activations::ActivationParameters |
| Hyperparameters for activations. More... | |
Namespaces | |
| namespace | FlexNN |
| Namespace for the FlexNN neural network library. | |
| namespace | FlexNN::Activations |
Variables | |
| constexpr double | FlexNN::Activations::kDefaultLeakyAlpha = 0.01 |
| Default slope for LeakyReLU negative side (matches LLD_FLEXNN ยง4, v0.1). | |
Per-activation hyperparameters for FlexNN::Activations.
Centralizes all activation-specific defaults so layers do not grow a new member per activation (avoids Dense::leakyAlpha, Dense::eluAlpha, ...). ActivationParameters is a trivial POD (8 B for v0.1) stored by-value in concrete layers (Dense, Conv1D) alongside the Activation enum. Layers pass it to detail::forward/backward; non-Leaky activations ignore it.
Future extensions (ELU alpha, Softmax temperature, etc.) add a field here with a constexpr default and update ModelIO aux handling without touching every layer's ctor signature.