Class Nade

A Neural Autoregressive Distribution Estimator (NADE).

Hierarchy

Nade

Index

Constructors

constructor

Properties

Methods

sample

Constructors

constructor

new Nade(encWeights: tf.Tensor3D, decWeightsT: tf.Tensor3D): Nade

- Defined in src/music_vae/model.ts:654
Nade contructor.

Parameters
- encWeights: tf.Tensor3D
  
  The encoder weights (kernel), sized [numDims, numHidden, 1].
- decWeightsT: tf.Tensor3D
  
  The transposed decoder weights (kernel), sized [numDims, numHidden, 1].
Returns Nade

Properties

decWeightsT

decWeightsT: tf.Tensor2D

encWeights

encWeights: tf.Tensor2D

numDims

numDims: number

numHidden

numHidden: number

Methods

sample

sample(encBias: tf.Tensor2D, decBias: tf.Tensor2D): tf.Tensor2D

- Defined in src/music_vae/model.ts:682
Samples from the NADE given a batch of encoder and decoder biases.

Selects the MAP (argmax) of each Bernoulli random variable.

Parameters
- encBias: tf.Tensor2D
  
  A batch of biases to use when encoding, sized [batchSize, numHidden].
- decBias: tf.Tensor2D
  
  A batch of biases to use when decoding, sized [batchSize, numDims].
Returns tf.Tensor2D

Generated using TypeDoc