dualbounds.iv.DualIVBounds.compute_dual_variables¶

DualIVBounds.compute_dual_variables(wprobs: ~numpy.array, ydists: list, ymin: float | None = None, ymax: float | None = None, min_quantile: float | None = None, interp_fn: callable = <function adaptive_interpolate>, nvals: int = 100, ninterp: int | None = None, verbose: bool = True, **kwargs)[source]¶

Same signature as generic.DualBounds.compute_dual_variables with the following exceptions.

Parameters:¶

wprobs : np.array

wprobs[i, z] is the estimated probability \(P(W(z) = i | X_i)\)

ydists : list

For z, w in {0,1}, ydists[z][w] must be a list of batched scipy dists whose shapes sum to n. The ith element of ydists[z][w] represents the law of \(Y_i(w) | W_i(z) = w, X_i\).

min_quantile : float

Minimum quantile to consider when discretizing. Defaults to 1 / (2*nvals).