import { DataTable } from '../data-table/data-table';
/**
 * Optimal 1D quantization using dynamic programming on a histogram.
 *
 * Pools all columns of the input DataTable into a single 1D dataset,
 * sorts the values, bins them using a blend of uniform and quantile
 * positioning, then uses DP to find k centroids that minimize weighted
 * sum-of-squared-errors (SSE).
 *
 * Bin positions are an adaptive blend of uniform (value-space) and
 * quantile (rank-space) positioning. The blend ratio is computed from
 * the data's IQR-to-range ratio: extreme outlier distributions (small
 * IQR relative to range) use near-pure quantile to give the dense
 * center adequate bins, while moderate-tail distributions reduce
 * quantile bias (but keep at least 50% quantile weighting).
 *
 * Bin weights use sub-linear density weighting: weight = count^alpha.
 * With alpha < 1, sparse tail regions earn meaningful influence on
 * centroid placement.
 *
 * @param dataTable - Input data table whose columns are pooled into 1D.
 * @param k - Number of codebook entries (default 256).
 * @param alpha - Density weight exponent. 0 = uniform (each bin equal),
 * 0.5 = sqrt (balanced), 1.0 = standard MSE (dense regions dominate).
 * Default 0.5.
 * @returns Object with `centroids` (DataTable with one 'data' column of
 * k Float32 values, sorted ascending) and `labels` (DataTable with same
 * column layout as input, each column containing Uint8Array indices into
 * the codebook).
 */
declare const quantize1d: (dataTable: DataTable, k?: number, alpha?: number) => {
    centroids: DataTable;
    labels: DataTable;
};
export { quantize1d };