import numpy import csb.test as test from csb.statmech.ensembles import BoltzmannEnsemble from csb.statmech.wham import WHAM, NonparametricWHAM class FunnyGaussian(object): def __init__(self, d, k=100.): self.d = int(d) self.k = float(k) def sample(self, n_samples, inv_T=1): from numpy.random import standard_normal from numpy import sqrt, sum from csb.statistics.rand import truncated_gamma x = standard_normal((self.d, n_samples)) x /= sqrt(sum(x ** 2, 0)) r = truncated_gamma(n_samples, 0.5 * self.d, self.k * inv_T, 0., 0.5) r = (2 * r) ** 0.5 return (x * r).T def energy(self, x): x = numpy.array(x) return 0.5 * self.k * numpy.sum(x ** 2, -1) def log_Z(self, beta=1.): from csb.numeric import log from scipy.special import gammainc, gammaln return log(0.5 * self.d) + log(gammainc(0.5 * self.d, 0.5 * self.k)) + \ gammaln(0.5 * self.d) + (0.5 * self.d) * (log(2) - log(self.k)) def log_g(self, energies): from csb.numeric import log return (0.5 * self.d - 1) * log(2 * energies / self.k) + log(self.d / self.k) @test.functional class TestWHAM(test.Case): def setUp(self): self.betas = numpy.linspace(1e-5, 1., 10) self.n = n = 1000 gaussian = FunnyGaussian(10, 100.) self.samples = [] self.raw_energies = [] for beta in self.betas: self.samples.append(gaussian.sample(n, beta)) self.raw_energies.append(gaussian.energy(self.samples[-1])) self.raw_energies = numpy.array(self.raw_energies) self.ensembles = [BoltzmannEnsemble(beta=beta) for beta in self.betas] self.log_z = gaussian.log_Z() self.log_g = gaussian.log_g(numpy.ravel(self.raw_energies)) def testWHAM(self): w = WHAM(self.ensembles, numpy.ravel(self.raw_energies), numpy.array([self.n] * len(self.betas))) w.estimate() self.assertAlmostEqual(numpy.dot(numpy.array([1, -1]), w.log_z(numpy.array([1., 0.]))), self.log_z, delta=0.5) def testNonparametricWHAM(self): w = NonparametricWHAM(self.ensembles, numpy.ravel(self.raw_energies), [self.n] * len(self.betas)) w.estimate() ens = [BoltzmannEnsemble(beta=1.,), BoltzmannEnsemble(beta=0.)] self.assertAlmostEqual(numpy.dot(numpy.array([1, -1]), w.log_z(ensembles=ens)), self.log_z, delta=0.5) if __name__ == '__main__': test.Console()