From 49a653081bea2dd3a8196f1184720928e258e20a Mon Sep 17 00:00:00 2001
From: Xiaoxu Guo <ftiasch0@gmail.com>
Date: Sat, 7 Oct 2017 15:42:15 +0800
Subject: [PATCH] added random state

---
 bayes_opt/bayesian_optimization.py | 20 +++++++++++++++-----
 bayes_opt/helpers.py               |  6 +++---
 2 files changed, 18 insertions(+), 8 deletions(-)

diff --git a/bayes_opt/bayesian_optimization.py b/bayes_opt/bayesian_optimization.py
index f497431f3..c28399d54 100644
--- a/bayes_opt/bayesian_optimization.py
+++ b/bayes_opt/bayesian_optimization.py
@@ -9,7 +9,7 @@
 
 class BayesianOptimization(object):
 
-    def __init__(self, f, pbounds, verbose=1):
+    def __init__(self, f, pbounds, random_state=None, verbose=1):
         """
         :param f:
             Function to be maximized.
@@ -25,6 +25,13 @@ def __init__(self, f, pbounds, verbose=1):
         # Store the original dictionary
         self.pbounds = pbounds
 
+        if random_state is None:
+            self.random_state = np.random.RandomState()
+        elif isinstance(random_state, int):
+            self.random_state = np.random.RandomState(random_state)
+        else:
+            self.random_state = random_state
+
         # Get the name of the parameters
         self.keys = list(pbounds.keys())
 
@@ -59,6 +66,7 @@ def __init__(self, f, pbounds, verbose=1):
         self.gp = GaussianProcessRegressor(
             kernel=Matern(nu=2.5),
             n_restarts_optimizer=25,
+            random_state=self.random_state
         )
 
         # Utility Function placeholder
@@ -87,7 +95,7 @@ def init(self, init_points):
         """
 
         # Generate random points
-        l = [np.random.uniform(x[0], x[1], size=init_points)
+        l = [self.random_state.uniform(x[0], x[1], size=init_points)
              for x in self.bounds]
 
         # Concatenate new random points to possible existing
@@ -276,7 +284,8 @@ def maximize(self,
         x_max = acq_max(ac=self.util.utility,
                         gp=self.gp,
                         y_max=y_max,
-                        bounds=self.bounds)
+                        bounds=self.bounds,
+                        random_state=self.random_state)
 
         # Print new header
         if self.verbose:
@@ -293,7 +302,7 @@ def maximize(self,
             pwarning = False
             if np.any((self.X - x_max).sum(axis=1) == 0):
 
-                x_max = np.random.uniform(self.bounds[:, 0],
+                x_max = self.random_state.uniform(self.bounds[:, 0],
                                           self.bounds[:, 1],
                                           size=self.bounds.shape[0])
 
@@ -315,7 +324,8 @@ def maximize(self,
             x_max = acq_max(ac=self.util.utility,
                             gp=self.gp,
                             y_max=y_max,
-                            bounds=self.bounds)
+                            bounds=self.bounds,
+                            random_state=self.random_state)
 
             # Print stuff
             if self.verbose:
diff --git a/bayes_opt/helpers.py b/bayes_opt/helpers.py
index aa8d0fb13..eb3159c27 100644
--- a/bayes_opt/helpers.py
+++ b/bayes_opt/helpers.py
@@ -6,7 +6,7 @@
 from scipy.optimize import minimize
 
 
-def acq_max(ac, gp, y_max, bounds):
+def acq_max(ac, gp, y_max, bounds, random_state):
     """
     A function to find the maximum of the acquisition function
 
@@ -35,14 +35,14 @@ def acq_max(ac, gp, y_max, bounds):
     """
 
     # Warm up with random points
-    x_tries = np.random.uniform(bounds[:, 0], bounds[:, 1],
+    x_tries = random_state.uniform(bounds[:, 0], bounds[:, 1],
                                  size=(100000, bounds.shape[0]))
     ys = ac(x_tries, gp=gp, y_max=y_max)
     x_max = x_tries[ys.argmax()]
     max_acq = ys.max()
 
     # Explore the parameter space more throughly
-    x_seeds = np.random.uniform(bounds[:, 0], bounds[:, 1],
+    x_seeds = random_state.uniform(bounds[:, 0], bounds[:, 1],
                                 size=(250, bounds.shape[0]))
     for x_try in x_seeds:
         # Find the minimum of minus the acquisition function