def continuous_induced_measure_ppf(var,
ab,
ii,
u_samples,
quad_tol=1e-8,
opt_tol=1e-6):
if is_bounded_continuous_variable(var):
# canonical domain of bounded variables does not coincide with scipy
# variable canonical pdf domain with loc,scale=0,1
lb, ub = -1, 1
loc, scale = -1, 2
else:
lb, ub = var.support()
loc, scale = 0, 1
shapes = get_distribution_info(var)[2]
# need to map x from canonical polynomial domain to canonical domain of pdf
pdf = lambda x: var.dist._pdf((x - loc) / scale, **shapes) / scale
def pdf(x):
vals = var.dist._pdf((x - loc) / scale, **shapes) / scale
#print('x',x,(x-loc)/scale,vals)
return vals
#pdf = var.pdf
#func = partial(continuous_induced_measure_cdf,pdf,ab,ii,lb,ub,quad_tol)
from pyapprox.cython.orthonormal_polynomials_1d import\
continuous_induced_measure_cdf_pyx
func = partial(continuous_induced_measure_cdf_pyx, pdf, ab, ii, lb,
quad_tol)
method = 'bisect'
samples = invert_monotone_function(func, [lb, ub], u_samples, method,
opt_tol)
assert np.all(np.isfinite(samples))
return samples
def univariate_quantile_continuous_variable(pdf, bounds, beta, opt_tol=1e-8,
quad_opts={}):
if quad_opts.get("epsabs", 1.49e-8) > opt_tol:
raise ValueError("epsabs must be smaller than opt_tol")
func = partial(univariate_cdf_continuous_variable,
pdf, bounds[0], bounds[1], quad_opts=quad_opts)
method = 'bisect'
quantile = invert_monotone_function(
func, bounds, np.array([beta]), method, opt_tol)
return quantile