def test_base_terms(name, args):
term = getattr(terms, name)(**args)
pyterm = getattr(pyterms, name)(**args)
compare_terms(term, pyterm)
compare_terms(terms.TermDiff(term), pyterms.TermDiff(pyterm))
compare_terms(terms.TermConvolution(term, 0.5),
pyterms.TermConvolution(pyterm, 0.5))
term0 = terms.SHOTerm(S0=1.0, w0=0.5, Q=1.5)
pyterm0 = pyterms.SHOTerm(S0=1.0, w0=0.5, Q=1.5)
compare_terms(term + term0, pyterm + pyterm0)
compare_terms(term * term0, pyterm * pyterm0)
term0 = terms.SHOTerm(S0=1.0, w0=0.5, Q=0.2)
pyterm0 = pyterms.SHOTerm(S0=1.0, w0=0.5, Q=0.2)
compare_terms(term + term0, pyterm + pyterm0)
compare_terms(term * term0, pyterm * pyterm0)
def test_citations(data):
pm = pytest.importorskip("pymc3")
x, diag, y, t = data
with pm.Model() as model:
term = terms.SHOTerm(S0=1.0, w0=0.5, Q=3.0)
gp = GaussianProcess(term, t=x, diag=diag)
gp.marginal("obs", observed=y)
assert model.__citations__["celerite2"] == CITATIONS
def test_marginal(data):
pm = pytest.importorskip("pymc3")
x, diag, y, t = data
with pm.Model() as model:
term = terms.SHOTerm(S0=1.0, w0=0.5, Q=3.0)
gp = GaussianProcess(term, t=x, diag=diag)
gp.marginal("obs", observed=y)
assert np.allclose(
model.fastfn(model.logpt)(model.test_point),
model.fastfn(gp.log_likelihood(y))(model.test_point),
)
def test_errors(data):
x, diag, y, t = data
term = terms.SHOTerm(S0=1.0, w0=0.5, Q=3.0)
gp = GaussianProcess(term)
# Need to call compute first
with pytest.raises(RuntimeError):
gp.log_likelihood(y)
# Sorted
with pytest.raises(AssertionError):
gp.compute(x[::-1], diag=diag)
gp._d.eval()
# 1D
with pytest.raises(ValueError):
gp.compute(np.tile(x[:, None], (1, 5)), diag=diag)
# Only one of diag and yerr
with pytest.raises(ValueError):
gp.compute(x, diag=diag, yerr=np.sqrt(diag))
# Not positive definite
with pytest.raises(celerite2.backprop.LinAlgError):
gp.compute(x, diag=-10 * diag)
gp._d.eval()
# Not positive definite with `quiet`
gp.compute(x, diag=-10 * diag, quiet=True)
ld = gp._log_det.eval()
assert np.isinf(ld)
assert ld < 0
# Compute correctly
gp.compute(x, diag=diag)
gp.log_likelihood(y).eval()
# Dimension mismatch
with pytest.raises(ValueError):
gp.log_likelihood(y[:-1]).eval()
with pytest.raises(ValueError):
gp.log_likelihood(np.tile(y[:, None], (1, 5)))
with pytest.raises(ValueError):
gp.predict(y, t=np.tile(t[:, None], (1, 5)))
# +
import pymc3 as pm
import celerite2.theano
from celerite2.theano import terms as theano_terms
with pm.Model() as model:
mean = pm.Normal("mean", mu=0.0, sigma=prior_sigma)
jitter = pm.Lognormal("jitter", mu=0.0, sigma=prior_sigma)
sigma1 = pm.Lognormal("sigma1", mu=0.0, sigma=prior_sigma)
rho1 = pm.Lognormal("rho1", mu=0.0, sigma=prior_sigma)
tau = pm.Lognormal("tau", mu=0.0, sigma=prior_sigma)
term1 = theano_terms.SHOTerm(sigma=sigma1, rho=rho1, tau=tau)
sigma2 = pm.Lognormal("sigma2", mu=0.0, sigma=prior_sigma)
rho2 = pm.Lognormal("rho2", mu=0.0, sigma=prior_sigma)
term2 = theano_terms.SHOTerm(sigma=sigma2, rho=rho2, Q=0.25)
kernel = term1 + term2
gp = celerite2.theano.GaussianProcess(kernel, mean=mean)
gp.compute(t, diag=yerr**2 + jitter, quiet=True)
gp.marginal("obs", observed=y)
pm.Deterministic("psd", kernel.get_psd(omega))
trace = pm.sample(
tune=1000,
draws=1000,
import pymc3 as pm
import celerite2.theano
from celerite2.theano import terms as theano_terms
with pm.Model() as model:
mean = pm.Normal("mean", mu=0.0, sigma=prior_sigma)
log_jitter = pm.Normal("log_jitter", mu=0.0, sigma=prior_sigma)
log_sigma1 = pm.Normal("log_sigma1", mu=0.0, sigma=prior_sigma)
log_rho1 = pm.Normal("log_rho1", mu=0.0, sigma=prior_sigma)
log_tau = pm.Normal("log_tau", mu=0.0, sigma=prior_sigma)
term1 = theano_terms.SHOTerm(
sigma=pm.math.exp(log_sigma1),
rho=pm.math.exp(log_rho1),
tau=pm.math.exp(log_tau),
)
log_sigma2 = pm.Normal("log_sigma2", mu=0.0, sigma=prior_sigma)
log_rho2 = pm.Normal("log_rho2", mu=0.0, sigma=prior_sigma)
term2 = theano_terms.SHOTerm(sigma=pm.math.exp(log_sigma2),
rho=pm.math.exp(log_rho2),
Q=0.25)
kernel = term1 + term2
gp = celerite2.theano.GaussianProcess(kernel, mean=mean)
gp.compute(t, diag=yerr**2 + pm.math.exp(log_jitter), quiet=True)
gp.marginal("obs", observed=y)
pm.Deterministic("psd", kernel.get_psd(omega))