LayerPropertyAttribute
Attributes · AiDotNet.Attributes
Declares architectural properties of a neural network layer for cataloging and test generation.
Inheritance: Attribute → LayerPropertyAttribute
Constructors
| Constructor | Summary |
|---|---|
LayerPropertyAttribute() |
Properties
| Property | Summary |
|---|---|
ComputeCost Cost { get; set; } |
Relative computational cost. |
LayerApiShape ApiShape { get; set; } |
The Forward method signature shape this layer uses. |
bool ChangesShape { get; set; } |
Whether the layer changes the shape of its input. |
bool HasTrainingMode { get; set; } |
Whether the layer behaves differently during training vs inference (e.g., Dropout, BatchNorm). |
bool IsStateful { get; set; } |
Whether the layer is stateful across forward passes (RNN hidden state, running mean). |
bool IsTrainable { get; set; } |
Whether the layer has trainable parameters (weights/biases). |
bool NormalizesInput { get; set; } |
Whether the layer normalizes input (LayerNorm, BatchNorm, etc.) so uniform-value inputs produce identical outputs regardless of the actual value. |
bool ProducesNonFiniteOutput { get; set; } |
Whether the layer legitimately produces ±Infinity values in its Forward output, by design rather than by numerical instability. |
bool SupportsBackpropagation { get; set; } |
Whether the layer supports backpropagation gradient computation. |
bool SupportsInPlace { get; set; } |
Whether the layer supports in-place operation (output can alias input). |
bool TrainsViaCustomLoss { get; set; } |
Whether the layer trains via a custom loss rather than gradients flowing from its Forward output. |
bool UsesSurrogateGradient { get; set; } |
Whether the layer uses surrogate gradients (e.g., spiking neural networks use a smooth approximation for the non-differentiable Heaviside step function). |
int ExpectedInputRank { get; set; } |
Expected input rank (number of dimensions). |
string TestConstructorArgs { get; set; } |
Constructor arguments as a comma-separated string for auto-generated tests. |
string TestInputShape { get; set; } |
The input tensor shape to use for auto-generated tests, as a comma-separated string. |
string TestSetupCode { get; set; } |
C# code to call on the layer after construction to set up graph data, adjacency matrices, or other domain-specific initialization required before Forward. |