def get_gamma_rates(a, YangGammaQuant, pp_gamma_ncat):
b = a
m = gdtrix(b, a, YangGammaQuant) # user defined categories
s = pp_gamma_ncat / sum(
m
) # multiplier to scale the so that the mean of the discrete distribution is one
return array(m) * s # SCALED VALUES
def get_q_rate_samples():
q_rates, alpha = res_q[np.random.choice(range(len(res_alpha))
), :], res_alpha[np.random.choice(
range(len(res_alpha)))]
if verbose: print("resample:", q_rates, alpha)
q_rates_bins = get_q_rates_time_bins(q_rates, q_shifts, time_bins)
rGamma = np.random.gamma(alpha, 1. / alpha, (10000, len(q_rates_bins) - 1))
q_rates_bins_G = rGamma * (q_rates_bins[:-1] * q_multi_A)
if verbose: print(q_rates_bins)
if verbose: print(np.mean(q_rates_bins_G, 0))
rho_bins_array = 1 - np.exp(-(q_rates_bins_G) * dTime)
rho_bins = np.mean(rho_bins_array, 0)
if verbose: print(1 - np.exp(-(q_rates_bins[:-1]) * dTime))
if verbose: print(rho_bins)
ncat = 100
YangGammaQuant = (np.linspace(0, 1, ncat + 1) -
np.linspace(0, 1, ncat + 1)[1] / 2)[1:]
rGamma = gdtrix(alpha, alpha, YangGammaQuant) # user defined categories
rGamma = np.repeat(rGamma,
len(q_rates_bins) - 1).reshape(
(ncat, len(q_rates_bins) - 1))
return q_rates_bins, rGamma, alpha
def gdtrix_comp(b, p):
return gdtrix(1.0, b, 1-p)
def gdtrix_(b, p):
return gdtrix(1.0, b, p)
def gdtrix_comp(b, p):
return gdtrix(1.0, b, 1-p)
def gdtrix_(b, p):
return gdtrix(1.0, b, p)
def _ppf(self, q, a):
return special.gdtrix(1.0, a, q)
def _ppf(self, q, a):
return special.gdtrix(1.0, a, q)
def quantile(self, p):
"""Return the p quantile of the Gamma posterior (using :func:`scipy.stats.gdtrix`).
- Used only by :class:`BayesUCB` and :class:`AdBandits` so far.
"""
return gdtrix(self.k, 1. / self.lmbda, p)
def gamma(self, shape, scale=1.0, size=None):
""" simulate Gamma variates """
U = self.uniform(0, 1, size)
X = special.gdtrix(1 / scale, shape, U)
return X
