def test_poisson_loglikelihood(self): l = 4.2 poisson = Poisson(l) samples = poisson.sample(100) loglik0 = stats.poisson.logpmf(samples, l) loglik1 = poisson.loglikelihood(samples) self.assertLess(max(abs(loglik0 - loglik1)), 1e-8)
def test_poisson_pickle(self):
tmp_file = mkstemp()[1]
p0 = Poisson(2.5)
with open(tmp_file, 'w') as handle:
dump({'p': p0}, handle)
with open(tmp_file) as handle:
p1 = load(handle)['p']
x = p0.sample(100)
self.assertLess(max(abs(p0.loglikelihood(x) - p1.loglikelihood(x))), 1e-6)
def test_poisson_sample(self): M = 10 N = 1000 l = 3.5 poisson = Poisson(l) # observed and expected frequencies f_obs = histogram(poisson.sample(N).ravel(), arange(M + 1) - 0.5)[0] f_exp = empty(M) for m in range(M - 1): f_exp[m] = stats.poisson.pmf(m, l) * N f_exp[M - 1] = (1. - stats.poisson.cdf(M - 2, l)) * N # Pearson's chi-squared test p = stats.chisquare(f_obs, f_exp)[1] # should fail once in 10000 tests, but not more often self.assertGreater(p, 0.0001)