@dabrush/schema

Shared TypeScript types and runtime validators for Distortion-aware brushing datasets.

Use this package when you need to check that raw HD/LD input or a preprocessed DAB dataset has the shape expected by @dabrush/preprocess-js, @dabrush/engine, and @dabrush/web.

Install From GitHub

yarn add github:distortion-aware-brushing/schema#main
# or
npm install github:distortion-aware-brushing/schema#main

The package is not published to npm yet. GitHub installation runs the package prepare script and builds dist/ locally during install.

Data Shapes

Raw input is just high-dimensional data and low-dimensional coordinates with aligned rows.

import type { DabRawInput } from "@dabrush/schema";

const raw: DabRawInput = {
  hd: [
    [0.1, 0.2, 0.3],
    [0.8, 0.1, 0.4]
  ],
  ld: [
    [120, 300],
    [180, 260]
  ],
  labels: ["a", "b"],
  features: ["f0", "f1", "f2"]
};

hd must be n x m. ld must be n x 2. This project does not create projections; PCA, UMAP, or t-SNE should happen before DAB preprocessing.

After preprocessing, a DabDataset adds SNN similarity in CSR format plus a KNN graph.

import type { DabDataset } from "@dabrush/schema";

type Dataset = DabDataset;

Similarity values are defined as “larger means closer.” In v0, similarity.format is always "csr" and similarity.metric is "snn".

Validation

import { assertRawInput, validateDabDataset, validateRawInput } from "@dabrush/schema";

const result = validateRawInput(raw);
if (!result.ok) {
  console.error(result.errors);
}

assertRawInput(raw); // throws when invalid

const datasetResult = validateDabDataset(dataset);

Use validate* when you want error lists, and assert* when throwing is the right control flow.

Exports

• DabRawInput
• DabDataset
• CsrSimilarity
• NumericMatrix
• validateRawInput(input)
• assertRawInput(input)
• validateDabDataset(dataset)
• assertDabDataset(dataset)
• DAB_SCHEMA_VERSION