ContextEncoder<T>
Models & Types · AiDotNet.NeuralNetworks.Tabular
Context encoder for TabR that processes retrieved neighbors into a context representation.
For Beginners
The context encoder does three things:
- Takes the similar samples found by retrieval
- Weights them by how similar they are (attention)
- Creates a single “context” representation that summarizes the neighbors
This context is then combined with the query to make the final prediction.
How It Works
The context encoder takes retrieved neighbors (values and labels) and combines them using attention-weighted aggregation and cross-attention with the query.
Constructors
| Constructor | Summary |
|---|---|
ContextEncoder(int embeddingDim, int labelDim, int numHeads) |
Initializes the context encoder. |
Properties
| Property | Summary |
|---|---|
long ParameterCount { get; } |
Gets the total parameter count. |
Methods
| Method | Summary |
|---|---|
Tensor<T> Forward(Tensor<T> queryEmbedding, RetrievalContext<T> context) |
Encodes the retrieved context using cross-attention. |
void ResetState() |
Resets internal state. |
void UpdateParameters(T learningRate) |
Updates parameters. |