def __init__(self, diagnostic): self.diagnostic = diagnostic self.probInfectedRVG = None # list of dirichlet distributions for moving out of the infected state self.probAlphaRVG = None # list of beta distributions for the alpha probability if the diagnostic test is NSB self.weeklyStateCostRVG = [ ] # list of gamma distributions for the annual cost of states self.singleStateCostRVG = [ ] # list of gamma distributions for the single cost of states # create Dirichlet distributions for transition probabilities self.probInfectedRVG = RVGs.Dirichlet(a=[ Data.P_INFECTED_CLEARED, Data.P_INFECTED_TBD, Data.P_INFECTED_TBM ]) # crate beta distribution for the alpha probability if self.diagnostic == Diagnostic.SOC: fit_output = MM.get_beta_params(mean=Data.P_DX_SOC, st_dev=Data.P_DX_SOC / 5) else: fit_output = MM.get_beta_params(mean=Data.P_DX_NSB, st_dev=Data.P_DX_NSB / 5) self.probAlphaRVG = RVGs.Beta(a=fit_output["a"], b=fit_output["b"]) # create gamma distributions for annual state cost for cost in Data.WEEKLY_STATE_COST: # if cost is zero, add a constant 0, otherwise add a gamma distribution if cost == 0: self.weeklyStateCostRVG.append(RVGs.Constant(value=0)) else: # find shape and scale of the assumed gamma distribution # no data available to estimate the standard deviation, so we assumed st_dev=cost / 5 fit_output = MM.get_gamma_params(mean=cost, st_dev=cost / 5) # append the distribution self.weeklyStateCostRVG.append( RVGs.Gamma(a=fit_output["a"], loc=0, scale=fit_output["scale"])) # create gamma distributions for single state cost if self.diagnostic == Diagnostic.SOC: single_cost = Data.SOC_ONE_TIME_COST else: single_cost = Data.NSB_ONE_TIME_COST for cost in single_cost: # if cost is zero, add a constant 0, otherwise add a gamma distribution if cost == 0: self.singleStateCostRVG.append(RVGs.Constant(value=0)) else: # find shape and scale of the assumed gamma distribution # no data available to estimate the standard deviation, so we assumed st_dev=cost / 5 fit_output = MM.get_gamma_params(mean=cost, st_dev=cost / 5) # append the distribution self.singleStateCostRVG.append( RVGs.Gamma(a=fit_output["a"], loc=0, scale=fit_output["scale"]))
def __init__(self, therapy): self.therapy = therapy self.probMatrixRVG = [ ] # list of dirichlet distributions for transition probabilities self.lnRelativeRiskRVG = None # normal distribution for the natural log of the treatment relative risk self.annualStateCostRVG = [ ] # list of gamma distributions for the annual cost of states self.annualStateUtilityRVG = [ ] # list of beta distributions for the annual utility of states # create Dirichlet distributions for transition probabilities j = 0 for prob in Data.PROB_MATRIX: self.probMatrixRVG.append(RVGs.Dirichlet(a=prob[j:])) j += 1 # treatment relative risk rr_ci = [1.2, 3] # confidence interval of the treatment relative risk # find the mean and st_dev of the normal distribution assumed for ln(RR) # sample mean ln(RR) mean_ln_rr = math.log(Data.TREATMENT_RR) # sample standard deviation of ln(RR) std_ln_rr = \ (math.log(rr_ci[1]) - math.log(rr_ci[0])) / (2 * stat.norm.ppf(1 - 0.05 / 2)) # create a normal distribution for ln(RR) self.lnRelativeRiskRVG = RVGs.Normal(loc=mean_ln_rr, scale=std_ln_rr) # create gamma distributions for annual state cost for cost in Data.ANNUAL_STATE_COST_NO: # if cost is zero, add a constant 0, otherwise add a gamma distribution if cost == 0: self.annualStateCostRVG.append(RVGs.Constant(value=0)) else: # find shape and scale of the assumed gamma distribution # no data available to estimate the standard deviation, so we assumed st_dev=cost / 5 fit_output = MM.get_gamma_params(mean=cost, st_dev=cost / 5) # append the distribution self.annualStateCostRVG.append( RVGs.Gamma(a=fit_output["a"], loc=0, scale=fit_output["scale"])) # create beta distributions for annual state utility for utility in Data.ANNUAL_STATE_UTILITY_0: # if utility is zero, add a constant 0, otherwise add a beta distribution if utility == 0: self.annualStateCostRVG.append(RVGs.Constant(value=0)) else: # find alpha and beta of the assumed beta distribution # no data available to estimate the standard deviation, so we assumed st_dev=cost / 4 fit_output = MM.get_beta_params(mean=utility, st_dev=utility / 4) # append the distribution self.annualStateUtilityRVG.append( RVGs.Beta(a=fit_output["a"], b=fit_output["b"]))
def test_dirichlet(rnd, a): # dirichlet random variate generator dirichlet_dist = RVGs.Dirichlet(a) # obtain samples samples = get_samples_multivariate(dirichlet_dist, rnd) # report mean and variance a0 = sum(a) if type(a) == list: a = np.array(a) mean = a * (1.0 / a0) var = np.zeros(len(a)) for i in range(len(a)): var[i] = (a[i] * (a0 - a[i])) / (((a0)**2) * (a0 + 1.0)) print_test_results_multivariate('Dirichlet', samples, expectation=mean, variance=var)