add tests

test/test_sam.py

+import pytest
+import numpy as np
+from sam import sam, covis, logprior, acf
+
+n = 4
+nsamples = 100000
+
+
+def generate_mucov(seed=0):
+  np.random.seed(seed)
+  mu = np.random.normal(0, 10, n)
+  cov = np.identity(n) + np.random.normal(0, 1e-2, n**2).reshape((n, n))
+  cov = (cov + cov.T) / 2
+  return (mu, cov)
+
+
+def logprob(x, x_mu, x_cov):
+  r = x - x_mu
+  return (-0.5 *
+    (r @ np.linalg.inv(x_cov) @ r + np.linalg.slogdet(2 * np.pi * x_cov)[1]))
+
+
+def test_sam():
+  mu, cov = generate_mucov()
+  x0 = np.zeros(n)
+  samples, _ = sam(x0, logprob, nsamples=nsamples, x_mu=mu, x_cov=cov)
+  samples = samples[nsamples // 4::10]
+  samp_mu = np.mean(samples, axis=0)
+  samp_cov = np.cov(samples, rowvar=False)
+
+  assert np.max(np.abs(samp_mu - mu)) < 0.1
+  assert np.max(np.abs(samp_cov - cov)) < 0.1
+
+
+def test_covis():
+  mu, cov = generate_mucov()
+  mu0 = mu + np.random.normal(0, 0.1, n)
+  cov0 = 0.9 * cov + np.random.normal(0, 1e-3, n**2).reshape((n, n))
+  cov0 = (cov0 + cov0.T) / 2
+
+  _, _, diags = covis(mu0,
+    cov0,
+    logprob,
+    nsamples=nsamples,
+    x_mu=mu,
+    x_cov=cov)
+
+  assert abs(diags['logevidence']) < 1e-3