LSTM | Wisty.js

An LSTM cell with a dense layer on its top.

Hierarchy

LSTM

Index

Constructors

constructor

Properties

Methods

Constructors

constructor

new LSTM(inputSize: number, hiddenSize: number, outputSize: number, dropout?: number): LSTM

- Defined in utils/lstm.ts:22
Parameters
- inputSize: number
  
  The dimension of the input data.
- hiddenSize: number
  
  The dimension of the output of the LSTM, passed to the dense layer.
- outputSize: number
  
  The dimension of the output data.
- Default value dropout: number = 0.2
  
  The dropout rate between the LSTM cell and the dense layer.
Returns LSTM

Properties

Private denseBias

denseBias: Tensor

Private denseWeights

denseWeights: Tensor

dropout

dropout: number

Private lstmBias

lstmBias: Tensor

Private lstmForgetBias

lstmForgetBias: Tensor

Private lstmInitC

lstmInitC: Tensor

Private lstmInitH

lstmInitH: Tensor

Private lstmKernel

lstmKernel: Tensor

Methods

export

export(): Promise<{}>

- Defined in utils/lstm.ts:110
Return all the LSTM model parameters.

Returns Promise<{}>

initLSTM

initLSTM(clone?: boolean): { c: tf.Tensor2D; h: tf.Tensor2D }

- Defined in utils/lstm.ts:48
Gives the initial state values of the LSTM (c and h).

Parameters
- Default value clone: boolean = true
  
  If it is necessary to clone states variable or no.
Returns { c: tf.Tensor2D; h: tf.Tensor2D }
- c: tf.Tensor2D
- h: tf.Tensor2D

load

load(weights: {}): void

- Defined in utils/lstm.ts:94
Update the given model parameters.

Parameters
- weights: {}
  - [key: string]: any
Returns void

predict

predict(x: tf.Tensor1D, c: tf.Tensor2D, h: tf.Tensor2D, mask?: tf.Tensor1D, temperature?: number): LSTMPrediction

- Defined in utils/lstm.ts:62
Make a prediction given an input and state values (c and h).

Parameters
- x: tf.Tensor1D
  
  A vector of shape [inputSize].
- c: tf.Tensor2D
  
  LSTM's state value.
- h: tf.Tensor2D
  
  LSTM's last output value.
- Optional mask: tf.Tensor1D
  
  A vector of ones and zeros of shape [outputSize].
- Default value temperature: number = 1
Returns LSTMPrediction

Generated using TypeDoc