Make metropolis_hastings seedable (#106)

* Make metropolis_hastings seedable * switching order of parameters so as to not break old code.
sgsellan · Dec 1, 2023 · 94ab93a · 94ab93a
1 parent 3570ae3
commit 94ab93a
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 7 deletions.
diff --git a/src/gpytoolbox/metropolis_hastings.py b/src/gpytoolbox/metropolis_hastings.py
@@ -1,7 +1,10 @@
 import numpy as np
-import random
 
-def metropolis_hastings(unnorm_distr, next_sample, x0 , num_samples=100):
+def metropolis_hastings(unnorm_distr,
+    next_sample,
+    x0,
+    num_samples=100,
+    rng=np.random.default_rng()):
     """Randomly sample according to an unnormalized distribution.
 
     Given a function which is proportional to a probabilistic density and a strategy for generating candidate points, returns a set of samples which will asymptotically tend to being a sample a random sample of the unknown distribution.
@@ -16,6 +19,8 @@ def metropolis_hastings(unnorm_distr, next_sample, x0 , num_samples=100):
         Initial sample
     num_samples : int
         Number of samples in output (this will be *more* than the total number of considered samples or evaluations of unnorm_distr)
+    rng : numpy rng, optional (default: new `np.random.default_rng()`)
+        which numpy random number generator to use
 
     Returns
     -------
@@ -59,7 +64,7 @@ def unnorm_distr(x):
         f1 = unnorm_distr(x1)
         #print(f1)
         # Generate random value between 0 and 1
-        r = random.uniform(0, 1)
+        r = rng.uniform(0, 1)
         #print(f1/f0)
         if r<(f1/f0):
             # If accepted, update current sample with candidate sample

diff --git a/test/test_metropolis_hastings.py b/test/test_metropolis_hastings.py
@@ -7,7 +7,7 @@
 
 class TestMetropolisHastings(unittest.TestCase):
     def test_analytic_1d(self):
-        np.random.seed(0)
+        rng = np.random.default_rng(4352)
         # 1D test
         # Sample next point from a normal distribution
         def next_sample(x0):
@@ -18,7 +18,7 @@ def next_sample(x0):
         def unnorm_distr(x):
             return np.max((1-np.abs(x[0]),1e-8))
 
-        S, F = gpytoolbox.metropolis_hastings(unnorm_distr,next_sample,np.array([0.1]),1000000)
+        S, F = gpytoolbox.metropolis_hastings(unnorm_distr,next_sample,np.array([0.1]),1000000,rng)
         # This should look like an absolute value pyramid function
         hist, bin_edges = np.histogram(S,bins=np.linspace(-1,1,101), density=True)
         bin_centers = (bin_edges[0:100] + bin_edges[1:101])/2.
@@ -30,7 +30,7 @@ def unnorm_distr(x):
         # plt.show(block=False)
 
     def test_analytic_2d(self):
-        np.random.seed(0)
+        rng = np.random.default_rng(8312)
         # plt.pause(10)
         # plt.close(plot1)
         # 2D test
@@ -42,7 +42,7 @@ def next_sample(x0):
         def unnorm_distr(x):
             return 100*multivariate_normal.pdf(x,mean=np.array([0.0,0.0]),cov=np.array([[0.01,0.0],[0.0,0.01]]))
 
-        S, F = gpytoolbox.metropolis_hastings(unnorm_distr,next_sample,np.array([0.01,0.01]),500000)
+        S, F = gpytoolbox.metropolis_hastings(unnorm_distr,next_sample,np.array([0.01,0.01]),500000,rng)
 
         nbins = 40
         H, xedges, yedges = np.histogram2d(S[:,0], S[:,1], density=True, bins=nbins)