Class Value

A scalar value that tracks its computation graph for automatic differentiation.

Because TypeScript has no operator overloading, arithmetic is performed via methods: a.add(b), a.mul(b), a.pow(n), etc.

Index

Constructors

Properties

Methods

add backward div exp log mul neg pow relu sub

Constructors

constructor

new Value(data: number, children?: Value[], localGrads?: number[]): Value
Parameters
- data: number
- children: Value[] = []
- localGrads: number[] = []
Returns Value
- Defined in autograd.ts:29

Properties

data

data: number

Scalar value computed during the forward pass.

grad

grad: number

Derivative of the loss with respect to this node, computed during backward.

Methods

add

add(other: number | Value): Value
Addition: this + other.
Parameters
- other: number | Value
Returns Value
- Defined in autograd.ts:37

backward

backward(): void
Backpropagate gradients through the computation graph rooted at this node. Sets this.grad = 1 and accumulates gradients on all ancestors.

Returns void
- Defined in autograd.ts:98

div

div(other: number | Value): Value
Division: this / other.
Parameters
- other: number | Value
Returns Value
- Defined in autograd.ts:89

exp

exp(): Value
Exponential (e^x).

Returns Value
- Defined in autograd.ts:63

log

log(): Value
Natural logarithm.

Returns Value
- Defined in autograd.ts:58

mul

mul(other: number | Value): Value
Multiplication: this * other.
Parameters
- other: number | Value
Returns Value
- Defined in autograd.ts:43

neg

neg(): Value
Negation: -this.

Returns Value
- Defined in autograd.ts:78

pow

pow(exponent: number): Value
Exponentiation: this ** exponent (exponent is a plain number).
Parameters
- exponent: number
Returns Value
- Defined in autograd.ts:49

relu

relu(): Value
Rectified linear unit: max(0, x).

Returns Value
- Defined in autograd.ts:69

sub

sub(other: number | Value): Value
Subtraction: this - other.
Parameters
- other: number | Value
Returns Value
- Defined in autograd.ts:83