@saehrimnir/druidjs / StressMDS

Class: StressMDS<T>

Defined in: dimred/StressMDS.js:50

Weighted metric MDS (stress majorization family)

Minimises σ(Y) = Σ_{i<j} w_ij ⋅ (‖y_i - y_j‖ - d_ij)² for a weighting you choose. An exponent q gives w_ij = d_ij^q, recovering the classical family: 0 is raw stress (SMACOF's objective), -1 Sammon stress (SAMMON's), -2 elastic scaling (KKMDS). A matrix or a function works too, where a zero weight drops the pair from the objective — the usual way to express missing dissimilarities.

It does not replace SMACOF or SAMMON: those minimise the same objectives with their own historical algorithms and reach different local minima.

Optimised by Jacobi-preconditioned gradient descent with a backtracking line search, warm-started from classical MDS. O(N²·d) per iteration.

Template

T

See

KKMDS for the weights: -2 preset.

Examples

// Sammon stress, converged further than SAMMON's own optimizer takes it
const Y = new StressMDS(X, { weights: -1 }).transform();

// Ignore pairs whose dissimilarity was never measured
const W = new Matrix(N, N, (i, j) => (observed(i, j) ? 1 : 0));
const Y = new StressMDS(D, { metric: "precomputed", weights: W }).transform();

Extends

• DR

Extended by

• KKMDS

Type Parameters

Type Parameter	Description
T extends InputType

Constructors

Constructor

new StressMDS<T>(X: T, parameters?: Partial<ParametersStressMDS>): StressMDS<T>;

Defined in: dimred/StressMDS.js:76

Weighted metric MDS.

Parameters

Parameter	Type	Description
X	T	The high-dimensional data, or a precomputed distance matrix.
parameters?	Partial<ParametersStressMDS>	Object containing parameterization of the DR method.

Returns

StressMDS<T>

Overrides

DR.constructor

Properties

Property	Type	Default value	Description	Inherited from	Defined in
__input	T	undefined	-	DR.__input	dimred/DR.js:46
_D	number	undefined	-	DR._D	dimred/DR.js:28
_energy	number	Infinity	-	-	dimred/StressMDS.js:63
_is_initialized	boolean	undefined	-	DR._is_initialized	dimred/DR.js:34
_N	number	undefined	-	DR._N	dimred/DR.js:30
_parameters	ParametersStressMDS	undefined	-	DR._parameters	dimred/DR.js:49
_randomizer	Randomizer	undefined	-	DR._randomizer	dimred/DR.js:32
_target_distances	Matrix | undefined	undefined	The target distances. Named apart from the base class's _D, which is the input dimensionality of X, not a matrix.	-	dimred/StressMDS.js:57
_type	"array" | "matrix" | "typed"	undefined	-	DR._type	dimred/DR.js:54
_weights	Matrix | undefined	undefined	-	-	dimred/StressMDS.js:60
X	Matrix	undefined	-	DR.X	dimred/DR.js:56
Y	Matrix	undefined	-	DR.Y	dimred/DR.js:58

Accessors

energy

Get Signature

get energy(): number;

Defined in: dimred/StressMDS.js:103

The weighted stress σ(Y) of the current embedding.

Returns

number

---

projection

Get Signature

get projection(): T;

Defined in: dimred/DR.js:219

Returns

T

The projection in the type of input X.

Inherited from

DR.projection

---

weights

Get Signature

get weights(): Matrix;

Defined in: dimred/StressMDS.js:113

The weight matrix actually in use, whatever form weights was given in.

Returns

Matrix

Methods

[dispose]()

dispose: void;

Defined in: dimred/DR.js:320

Alias of release for the using declaration, so a hand-driven run frees its buffers when the block exits. Note that transform and a completed or break-ed generator() already release on their own — this only matters for a generator abandoned part way.

using tsne = new TSNE(X, { d: 2 });
const steps = tsne.generator(500);
steps.next(); // buffers released when the enclosing block exits

Returns

void

Inherited from

DR.[dispose]

---

check_init()

check_init(): DR<T, ParametersStressMDS>;

Defined in: dimred/DR.js:210

If the respective DR method has an init function, call it before transform.

Returns

DR<T, ParametersStressMDS>

Inherited from

DR.check_init

---

generator()

generator(): Generator<T, T, void>;

Defined in: dimred/StressMDS.js:326

Computes the projection.

Returns

Generator<T, T, void>

A generator yielding the intermediate steps of the projection.

Overrides

DR.generator

---

init()

init(): StressMDS<T>;

Defined in: dimred/StressMDS.js:186

Computes the target distances, the weights, and the starting configuration.

Returns

StressMDS<T>

Overrides

DR.init

---

parameter()

Call Signature

parameter(): ParametersStressMDS;

Defined in: dimred/DR.js:82

Get all Parameters.

Returns

ParametersStressMDS

Inherited from

DR.parameter

Call Signature

parameter<K>(name: K): ParametersStressMDS[K];

Defined in: dimred/DR.js:88

Get value of given parameter.

Type Parameters

Type Parameter	Description
K extends keyof ParametersStressMDS

Parameters

Parameter	Type	Description
name	K	Name of the parameter.

Returns

ParametersStressMDS[K]

Inherited from

DR.parameter

Call Signature

parameter<K>(name: K, value: ParametersStressMDS[K]): StressMDS<T>;

Defined in: dimred/DR.js:95

Set value of given parameter.

Type Parameters

Type Parameter	Description
K extends keyof ParametersStressMDS

Parameters

Parameter	Type	Description
name	K	Name of the parameter.
value	ParametersStressMDS[K]	Value of the parameter to set.

Returns

StressMDS<T>

Inherited from

DR.parameter

---

release()

release(): StressMDS<T>;

Defined in: dimred/DR.js:302

Hands back the WASM buffers this instance is holding.

Only needed after driving generator() by hand and stopping early — a plain transform(), or a for…of over generator() (including one you break), already releases when it ends. It frees only this method's buffers, never another running instance's, and the next run simply reallocates, so it is safe to call at any time, more than once, and while other instances are mid-run.

Returns

StressMDS<T>

Example

const tsne = new TSNE(X, { d: 2 });
const steps = tsne.generator(500);
steps.next();
tsne.release(); // stop early and give the buffers back

Inherited from

DR.release

---

transform()

transform(): T;

Defined in: dimred/StressMDS.js:386

Computes the projection.

Returns

T

Overrides

DR.transform

---

transform_async()

transform_async(...args: unknown[]): Promise<T>;

Defined in: dimred/DR.js:241

Computes the projection.

Parameters

Parameter	Type	Description
...args	unknown[]	Arguments the transform method of the respective DR method takes.

Returns

Promise<T>

The dimensionality reduced dataset.

Inherited from

DR.transform_async

---

generator()

static generator<T>(X: T, parameters?: Partial<ParametersStressMDS>): Generator<T, T, void>;

Defined in: dimred/StressMDS.js:412

Type Parameters

Type Parameter	Description
T extends InputType

Parameters

Parameter	Type	Description
X	T	-
parameters?	Partial<ParametersStressMDS>	-

Returns

Generator<T, T, void>

Overrides

DR.generator

---

transform()

static transform<T>(X: T, parameters?: Partial<ParametersStressMDS>): T;

Defined in: dimred/StressMDS.js:401

Type Parameters

Type Parameter	Description
T extends InputType

Parameters

Parameter	Type	Description
X	T	-
parameters?	Partial<ParametersStressMDS>	-

Returns

T

Overrides

DR.transform

---

transform_async()

static transform_async<T>(X: T, parameters?: Partial<ParametersStressMDS>): Promise<T>;

Defined in: dimred/StressMDS.js:424

Type Parameters

Type Parameter	Description
T extends InputType

Parameters

Parameter	Type	Description
X	T	-
parameters?	Partial<ParametersStressMDS>	-

Returns

Promise<T>

Overrides

DR.transform_async