| 1 |
import numpy |
| 2 |
import csb.test as test |
| 3 |
|
| 4 |
from csb.statmech.ensembles import BoltzmannEnsemble |
| 5 |
from csb.statmech.wham import WHAM, NonparametricWHAM |
| 6 |
|
| 7 |
|
| 8 |
class FunnyGaussian(object): |
| 9 |
|
| 10 |
def __init__(self, d, k=100.): |
| 11 |
|
| 12 |
self.d = int(d) |
| 13 |
self.k = float(k) |
| 14 |
|
| 15 |
def sample(self, n_samples, inv_T=1): |
| 16 |
|
| 17 |
from numpy.random import standard_normal |
| 18 |
from numpy import sqrt, sum |
| 19 |
from csb.statistics.rand import truncated_gamma |
| 20 |
|
| 21 |
x = standard_normal((self.d, n_samples)) |
| 22 |
x /= sqrt(sum(x ** 2, 0)) |
| 23 |
|
| 24 |
r = truncated_gamma(n_samples, 0.5 * self.d, self.k * inv_T, 0., 0.5) |
| 25 |
r = (2 * r) ** 0.5 |
| 26 |
|
| 27 |
return (x * r).T |
| 28 |
|
| 29 |
def energy(self, x): |
| 30 |
|
| 31 |
x = numpy.array(x) |
| 32 |
return 0.5 * self.k * numpy.sum(x ** 2, -1) |
| 33 |
|
| 34 |
def log_Z(self, beta=1.): |
| 35 |
|
| 36 |
from csb.numeric import log |
| 37 |
from scipy.special import gammainc, gammaln |
| 38 |
|
| 39 |
return log(0.5 * self.d) + log(gammainc(0.5 * self.d, 0.5 * self.k)) + \ |
| 40 |
gammaln(0.5 * self.d) + (0.5 * self.d) * (log(2) - log(self.k)) |
| 41 |
|
| 42 |
def log_g(self, energies): |
| 43 |
|
| 44 |
from csb.numeric import log |
| 45 |
return (0.5 * self.d - 1) * log(2 * energies / self.k) + log(self.d / self.k) |
| 46 |
|
| 47 |
|
| 48 |
@test.functional |
| 49 |
class TestWHAM(test.Case): |
| 50 |
|
| 51 |
def setUp(self): |
| 52 |
self.betas = numpy.linspace(1e-5, 1., 10) |
| 53 |
self.n = n = 1000 |
| 54 |
|
| 55 |
gaussian = FunnyGaussian(10, 100.) |
| 56 |
|
| 57 |
self.samples = [] |
| 58 |
self.raw_energies = [] |
| 59 |
|
| 60 |
|
| 61 |
for beta in self.betas: |
| 62 |
self.samples.append(gaussian.sample(n, beta)) |
| 63 |
self.raw_energies.append(gaussian.energy(self.samples[-1])) |
| 64 |
|
| 65 |
self.raw_energies = numpy.array(self.raw_energies) |
| 66 |
self.ensembles = [BoltzmannEnsemble(beta=beta) for beta in self.betas] |
| 67 |
|
| 68 |
self.log_z = gaussian.log_Z() |
| 69 |
self.log_g = gaussian.log_g(numpy.ravel(self.raw_energies)) |
| 70 |
|
| 71 |
def testWHAM(self): |
| 72 |
|
| 73 |
w = WHAM(self.ensembles, |
| 74 |
numpy.ravel(self.raw_energies), |
| 75 |
numpy.array([self.n] * len(self.betas))) |
| 76 |
w.estimate() |
| 77 |
|
| 78 |
self.assertAlmostEqual(numpy.dot(numpy.array([1, -1]), |
| 79 |
w.log_z(numpy.array([1., 0.]))), |
| 80 |
self.log_z, delta=0.5) |
| 81 |
|
| 82 |
def testNonparametricWHAM(self): |
| 83 |
|
| 84 |
w = NonparametricWHAM(self.ensembles, |
| 85 |
numpy.ravel(self.raw_energies), |
| 86 |
[self.n] * len(self.betas)) |
| 87 |
w.estimate() |
| 88 |
ens = [BoltzmannEnsemble(beta=1.,), |
| 89 |
BoltzmannEnsemble(beta=0.)] |
| 90 |
self.assertAlmostEqual(numpy.dot(numpy.array([1, -1]), |
| 91 |
w.log_z(ensembles=ens)), |
| 92 |
self.log_z, delta=0.5) |
| 93 |
|
| 94 |
|
| 95 |
if __name__ == '__main__': |
| 96 |
|
| 97 |
test.Console() |
| 98 |
|