RetrievalModule<T>
Models & Types · AiDotNet.NeuralNetworks.Tabular
Retrieval module for TabR (Retrieval-Augmented Tabular Learning).
For Beginners
The retrieval module works like a smart lookup:
- For each test sample, find similar training samples
- Retrieve their features and labels
- Use this information to help make better predictions
Think of it as “looking up similar past cases” before making a decision.
How It Works
The retrieval module finds similar training examples to a query sample and provides their features and labels as additional context for prediction. This is similar to k-NN but integrated into a neural network.
Constructors
| Constructor | Summary |
|---|---|
RetrievalModule(int embeddingDim, int numNeighbors, double temperature) |
Initializes the retrieval module. |
Properties
| Property | Summary |
|---|---|
bool HasTrainingData { get; } |
Gets whether training data has been stored. |
int NumNeighbors { get; } |
Gets the number of neighbors to retrieve. |
Methods
| Method | Summary |
|---|---|
RetrievalContext<T> Retrieve(Tensor<T> queryKeys) |
Retrieves similar samples for a batch of queries. |
void ClearTrainingData() |
Clears stored training data. |
void StoreTrainingData(Tensor<T> keys, Tensor<T> values, Tensor<T> labels) |
Stores training data for retrieval. |