import {Vol} from 'convnet-sourcecode/src/convnet_vol.ts'

public class | source

Vol

Store a 3D volume of numbers with width(sx), height(sy), depth(depth) for the network. It also holds weights and gradients w.r.t the weights.

Constructor Summary

Public Constructor
public	constructor()

Member Summary

Public Members
public	depth: *
public	dw: *
public	sx: number
public	sy: *
public	w: *

Method Summary

Public Methods
public	add(x: number, y: number, d: number, v: number) Address the node and add the given `value`(v) to its weight.
public	addFrom(volume: Vol) Add the weights of a given network `volume` to current network
public	addFromScaled(volume: Vol, alpha: number) Add the weights of a given network `volume` with scale `alpha` to current network
public	add_grad(x: number, y: number, d: number, v: number) Address the node and add the given `value`(v) to its gradient.
public	clone(): Vol Return a clone of current network with the same parameters, only reset gradients to 0
public	cloneAndZero(): Vol Return a clone of current network and initialize the weights and gradients as 0
public	fromJSON(obj: VolInterface) Reset the netwrok information parameters informations
public	get(x: number, y: number, d: number): number Address the node and return its weight.
public	getNodeIndex(x: number, y: number, d: number): number Address the node by given `width`(x), `height`(y) and `depth`(d), then return the index of this node in the flatten 1D list.
public	get_grad(x: number, y: number, d: number): number Address the node and return its gradient.
public	init(sx: undefined, sy: undefined, depth: number, c: number) Initialize the network input placeholders and parameters If the given `width` is an array, we assume that 1D volume is provided If the given `width`/`height`/`depth` are all `number`, it will be treated as a 3D vlumn
public	set(x: number, y: number, d: number, v: number) Address the node and reset its weight to the given `value`(v).
public	setConst(value: number) Reset the weights of current network to a given constant number `value`
public	set_grad(x: number, y: number, d: number, v: number) Address the node and reset its gradient to the given `value`(v).
public	toJSON(): VolInterface Expose the netwrok parameters information

Public Constructors

public constructor() source

Public Members

public depth: * source

public dw: * source

public sx: number source

public sy: * source

public w: * source

Public Methods

public add(x: number, y: number, d: number, v: number) source

Address the node and add the given value(v) to its weight.

Params:

Name	Type	Attribute	Description
x	number
y	number
d	number
v	number

public addFrom(volume: Vol) source

Add the weights of a given network volume to current network

Params:

Name	Type	Attribute	Description
volume	Vol

public addFromScaled(volume: Vol, alpha: number) source

Add the weights of a given network volume with scale alpha to current network

Params:

Name	Type	Attribute	Description
volume	Vol
alpha	number

public add_grad(x: number, y: number, d: number, v: number) source

Address the node and add the given value(v) to its gradient.

Params:

Name	Type	Attribute	Description
x	number
y	number
d	number
v	number

public clone(): Vol source

Return a clone of current network with the same parameters, only reset gradients to 0

Return:

Vol

public cloneAndZero(): Vol source

Return a clone of current network and initialize the weights and gradients as 0

Return:

Vol

public fromJSON(obj: VolInterface) source

Reset the netwrok information parameters informations

Params:

Name	Type	Attribute	Description
obj	VolInterface

public get(x: number, y: number, d: number): number source

Address the node and return its weight.

Params:

Name	Type	Attribute	Description
x	number
y	number
d	number

Return:

number

public getNodeIndex(x: number, y: number, d: number): number source

Address the node by given width(x), height(y) and depth(d), then return the index of this node in the flatten 1D list.

Params:

Name	Type	Attribute	Description
x	number
y	number
d	number

Return:

number

public get_grad(x: number, y: number, d: number): number source

Address the node and return its gradient.

Params:

Name	Type	Attribute	Description
x	number
y	number
d	number

Return:

number

public init(sx: undefined, sy: undefined, depth: number, c: number) source

Initialize the network input placeholders and parameters

If the given width is an array, we assume that 1D volume is provided
If the given width/height/depth are all number, it will be treated as a 3D vlumn

Params:

Name	Type	Attribute	Description
sx	undefined
sy	undefined
depth	number
c	number

public set(x: number, y: number, d: number, v: number) source

Address the node and reset its weight to the given value(v).

Params:

Name	Type	Attribute	Description
x	number
y	number
d	number
v	number

public setConst(value: number) source

Reset the weights of current network to a given constant number value

Params:

Name	Type	Attribute	Description
value	number

public set_grad(x: number, y: number, d: number, v: number) source

Address the node and reset its gradient to the given value(v).

Params:

Name	Type	Attribute	Description
x	number
y	number
d	number
v	number

public toJSON(): VolInterface source

Expose the netwrok parameters information

Return:

VolInterface