Class HierarchicalEncoder

A hierarchical encoder that uses the outputs from each level as the inputs to the subsequent level.

Hierarchy

Encoder
- HierarchicalEncoder

Index

Constructors

constructor

Properties

Methods

encode

Constructors

constructor

new HierarchicalEncoder(baseEncoders: Encoder[], numSteps: number[], muVars: LayerVars): HierarchicalEncoder

- Defined in music_vae/model.ts:154
HierarchicalEncoder contructor.

Parameters
- baseEncoders: Encoder[]
  
  An list of Encoder objects to use for each.
- numSteps: number[]
  
  A list containing the number of steps (outputs) for each level of the hierarchy. This number should evenly divide the inputs for each level. The final entry must always be 1.
- muVars: LayerVars
  
  The LayerVars for projecting from the final states of the final level to the mean mu of the random variable, z.
Returns HierarchicalEncoder

Properties

baseEncoders

baseEncoders: Encoder[]

muVars

muVars: LayerVars

numSteps

numSteps: number[]

zDims

zDims: number

Methods

encode

encode(sequence: tf.Tensor3D, segmentLengths?: number[]): Tensor<R2>

Overrides Encoder.encode
- Defined in music_vae/model.ts:181
Encodes a batch of sequences.

Parameters
- sequence: tf.Tensor3D
  
  The batch of sequences to be encoded.
- Optional segmentLengths: number[]
  
  (Optional) An array of lengths of the base-level segments. Must have length numSteps[0]. Assumes that batch size is 1.
Returns Tensor<R2>

A batch of mu values.

Generated using TypeDoc