Merge pull request guillermo-navas-palencia#2 from alonme/support-ver…

…y-large-beta-numbers

Support very large beta numbers

cprior/cdist/beta.py

@@ -22,6 +22,11 @@
 from .utils import check_ab_method
 from .utils import check_mv_method
 
+def get_integration_points(a, b):
+ if a == 1 and b == 1:
+ return None
+
+ return [(a-1)/(a+b-2)]
 
 def func_ppf(x, a0, b0, a1, b1, p):
  """Function CDF ratio of beta function for root-finding."""
@@ -796,7 +801,9 @@ def probability_vs_all(self, method="quad", variant="B", lift=0,
  a = self.models[variant].alpha_posterior
  b = self.models[variant].beta_posterior
 
- return integrate.quad(func=func_mv_prob, a=0, b=1, args=(
+ points = get_integration_points(a, b)
+
+ return integrate.quad(func=func_mv_prob, a=0, b=1, points=points, args=(
  a, b, variant_params))[0]
  else:
  # prepare parameters
@@ -1031,18 +1038,20 @@ def expected_loss_relative_vs_all(self, method="quad", control="A",
 
  return (maxall / xvariant).mean() - 1
  else:
+ a = self.models[variant].alpha_posterior
+ b = self.models[variant].beta_posterior
+
  if method == "quad":
  variant_params = [(self.models[v].alpha_posterior,
  self.models[v].beta_posterior)
  for v in variants]
 
- e_max = integrate.quad(func=func_mv_elr, a=0, b=1, args=(
+ points = get_integration_points(a, b)
+ e_max = integrate.quad(func=func_mv_elr, a=0, b=1, points=points, args=(
  variant_params))[0]
  else:
  e_max = self._expected_value_max_mlhs(variants, mlhs_samples)
 
- a = self.models[variant].alpha_posterior
- b = self.models[variant].beta_posterior
  e_inv_x = (a + b - 1) / (a - 1)
 
  return e_max * e_inv_x - 1
@@ -1180,8 +1189,10 @@ def expected_loss_vs_all(self, method="quad", variant="B", lift=0,
  b = self.models[variant].beta_posterior
 
  if method == "quad":
- return integrate.quad(func=func_mv_el, a=0, b=1, args=(
+ points = get_integration_points(a, b)
+ return integrate.quad(func=func_mv_el, a=0, b=1, points=points, args=(
  a, b, variant_params))[0]
+
  else:
  r = np.arange(1, mlhs_samples + 1)
  np.random.shuffle(r)