jaxopt.isotonic.isotonic_l2_pav

jaxopt.isotonic.isotonic_l2_pav(y, y_min=-inf, y_max=inf, increasing=True)[source]

Solves an isotonic regression problem using PAV.

Parameters
  • y – input to isotonic regression, a 1d-array.

  • y_min – Lower bound on the lowest predicted value.

  • y_max – Upper bound on the highest predicted value

  • increasing – Order of the constraints: If True, it solves \(\mathop{\mathrm{arg\,min}}_{v_1 \leq ... \leq v_n} \|v - y\|^2\). If False, it solves \(\mathop{\mathrm{arg\,min}}_{v_1 \geq ... \geq v_n} \|v - y\|^2\).

Returns

The solution, an array of the same size as y.