def CONDMAX(*arg):
""" Conditional maximum function.
Conputes a rv X, consisting of the maximum of performance
if corresponding num is greater than 0 for each MC trial.
Args:
arg: Sympy parsed arguments, [n0, p0, n1, p1, ...]
each n, p can be either a distribution or a scalar.
Return:
x: result distribution.
"""
ns = arg[::2]
ps = arg[1::2]
assert ns and ps and len(ns) == len(ps)
if not __has_distribution(ns) and not __has_distribution(ps):
candidate = [p for n, p in zip(ns, ps) if n > 0]
return max(candidate)
else:
x = Distribution.GetDummy()
mcpts = [None] * mcp.npts
for i in range(mcp.npts):
candidate = [
p._mcpts[i] if isinstance(p, UncertainFunction) else p
for n, p in zip(ns, ps)
if (n._mcpts[i] if isinstance(n, UncertainFunction) else n) > 0
]
mcpts[i] = max(
candidate) if candidate else Distribution.NON_ZERO_FACTOR
x._mcpts = np.asarray(mcpts)
return x
def pollack_boxcox_uncertainty(d, perc):
mean = get_pollack(d)
std = mean * perc
gt_perf = pollack_uncertainty(d, perc)
if std:
return Distribution.DistributionFromBoxCoxGaussian(mean, std, gt_perf._mcpts, lower=0)
else:
return gt_perf
def fabrication_boxcox_uncertainty(d, n):
if n > 0:
func = UncertaintyModel.fabrication()
gt = func(d, n)
tgt_mean = gt.mean
tgt_std = gt.var ** .5
return Distribution.DistributionFromBoxCoxGaussian(tgt_mean, tgt_std, gt._mcpts, lower=0)
else:
return 0
def design_pollack_boxcox_uncertainty(d, p0, perc):
mean = get_pollack(d)
std = mean * perc
gt_perf = design_pollack_uncertainty(d, p0, perc)
if std:
tgt_mean = gt_perf.mean
tgt_std = gt_perf.var ** .5
return Distribution.DistributionFromBoxCoxGaussian(tgt_mean, tgt_std, gt_perf._mcpts, lower=0)
else:
return gt_perf
def design_pollack_gaussian_uncertainty(d, p0, perc, a=None, b=None):
mean = get_pollack(d)
std = perc * mean
return Distribution.QGaussianDistribution(p0, mean, std, a, b)
def design_pollack_uncertainty(d, p0, perc):
mean = get_pollack(d)
std = perc * mean
return Distribution.QLogNormalDistribution(p0, mean, std)
def pollack_uncertainty(d, perc):
mean = get_pollack(d)
std = perc * mean
return Distribution.LogNormalDistribution(mean, std)
def higher_order_bernoulli(d, num, func):
return Distribution.HigherOrderBernoulli(func(d), num)