Skip to content

@saehrimnir/druidjs / ParametersStressMDS

Interface: ParametersStressMDS

Defined in: dimred/index.js:71

Properties

PropertyTypeDescriptionDefined in
d?numberthe dimensionality of the projection.dimred/index.js:72
eig_args?Partial<EigenArgs>Parameters for the eigendecomposition algorithm.dimred/index.js:86
epsilon?numberstop once the relative stress improvement falls below this.dimred/index.js:77
init_DR?"MDS" | "PCA" | "random"starting configuration. "MDS" runs classical MDS on the same distances, which is what keeps the non-convex descent out of the poor local minima this objective is known for. "PCA" needs the original data, not a precomputed matrix.dimred/index.js:81
iterations?numbermaximum number of gradient steps.dimred/index.js:76
learning_rate?numberinitial step size. Dimensionless: the gradient is preconditioned by the weighted degree, so this needs no rescaling for the data or the weighting. Adapted by the line search, so it only sets where the search starts.dimred/index.js:78
metric?Metric | "precomputed"the metric which defines the distance between two points. Pass graph shortest-path distances as "precomputed" for a graph layout.dimred/index.js:73
seed?numberthe seed for the random number generator.dimred/index.js:85
weights?WeightSpecPair weighting, see WeightSpec.dimred/index.js:75