def __init__(self, t, y, dy, p=2, q=1):
self._t = t.copy()
self._y = y.copy()
self._dy = dy.copy()
self._p = p
self._q = q
self._tv = cm.vecD()
self._tv.extend(t)
self._yv = cm.vecD()
self._yv.extend(y)
self._dyv = cm.vecD()
self._dyv.extend(dy)
self._carma_process = cm.run_mcmc_carma(1, 1, self._tv, self._yv,
self._dyv, self.p, self.q, 1,
False, 1)
self._carma_process.SetMLE(True)
self._T = self.t[-1] - self.t[0]
self._dt_min = np.min(np.diff(self.t))
self._mean_variance()
def doit(args):
pModel = int(args[0])
x, y, dy = args[1]
nSample = 10000
nBurnin = 1000
nThin = 1
nWalkers = 10
# Should not have to do this...
xv = carmcmc.vecD()
xv.extend(x)
yv = carmcmc.vecD()
yv.extend(y)
dyv = carmcmc.vecD()
dyv.extend(dy)
if pModel == 1:
sampler = carmcmc.run_mcmc_car1(nSample, nBurnin, xv, yv, dyv, nWalkers, nThin)
samplep = carmcmc.CarSample1(x, y, dy, sampler)
else:
sampler = carmcmc.run_mcmc_carma(nSample, nBurnin, xv, yv, dyv, pModel, 0, nWalkers, False, nThin)
samplep = carmcmc.CarmaSample(x, y, dy, sampler)
dic = samplep.DIC()
print "DIC", pModel, dic
return samplep
def setUp(self):
self.nSample = 100
self.nBurnin = 10
self.nThin = 1
self.nWalkers = 2
self.xdata = carmcmc.vecD()
self.xdata.extend(1.0 * np.arange(10))
self.ydata = carmcmc.vecD()
self.ydata.extend(1.0 * np.random.random(10))
self.dydata = carmcmc.vecD()
self.dydata.extend(0.1 * np.random.random(10))
def setUp(self):
self.nSample = 100
self.nBurnin = 10
self.nThin = 1
self.nWalkers = 2
self.xdata = carmcmc.vecD()
self.xdata.extend(1.0 * np.arange(10))
self.ydata = carmcmc.vecD()
self.ydata.extend(1.0 * np.random.random(10))
self.dydata = carmcmc.vecD()
self.dydata.extend(0.1 * np.random.random(10))
def testKalmanp(self):
pModel = 4
qModel = 0
sampler = carmcmc.run_mcmc_carma(self.nSample, self.nBurnin,
self.xdata, self.ydata, self.dydata,
pModel, qModel, self.nWalkers, False,
self.nThin)
psampler = carmcmc.CarmaSample(np.array(self.xdata),
np.array(self.ydata),
np.array(self.dydata), sampler)
sigsqr = (psampler._samples["sigma"][0]**2)[0]
ma_coefs = carmcmc.vecD()
ma_coefs0 = psampler._samples["ma_coefs"][0]
if len(ma_coefs0) != pModel:
ma_coefs0 = np.append(ma_coefs0, np.zeros(pModel - qModel - 1))
ma_coefs.extend(ma_coefs0)
omega = carmcmc.vecC()
for i in range(psampler.p):
omega.append(psampler._samples["ar_roots"][0][i])
#import pdb; pdb.set_trace()
kfilter = carmcmc.KalmanFilterp(self.xdata, self.ydata, self.dydata,
sigsqr, omega, ma_coefs)
kfilter.Filter()
pred0 = kfilter.Predict(self.xdata[0]) # evaluate at data point
val0 = pred0.first
var0 = pred0.second
predN = kfilter.Predict(self.xdata[-1] + 1) # extrapolate
valN = predN.first
varN = predN.second
self.assertTrue(varN > var0)
def testKalmanp(self):
pModel = 4
qModel = 0
sampler = carmcmc.run_mcmc_carma(self.nSample, self.nBurnin,
self.xdata, self.ydata, self.dydata,
pModel, qModel, self.nWalkers, False, self.nThin)
psampler = carmcmc.CarmaSample(np.array(self.xdata), np.array(self.ydata), np.array(self.dydata), sampler)
sigsqr = (psampler._samples["sigma"][0]**2)[0]
ma_coefs = carmcmc.vecD()
ma_coefs0 = psampler._samples["ma_coefs"][0]
if len(ma_coefs0) != pModel:
ma_coefs0 = np.append(ma_coefs0, np.zeros(pModel - qModel - 1))
ma_coefs.extend(ma_coefs0)
omega = carmcmc.vecC()
for i in range(psampler.p):
omega.append(psampler._samples["ar_roots"][0][i])
#import pdb; pdb.set_trace()
kfilter = carmcmc.KalmanFilterp(self.xdata, self.ydata, self.dydata, sigsqr, omega, ma_coefs)
kfilter.Filter()
pred0 = kfilter.Predict(self.xdata[0]) # evaluate at data point
val0 = pred0.first
var0 = pred0.second
predN = kfilter.Predict(self.xdata[-1]+1) # extrapolate
valN = predN.first
varN = predN.second
self.assertTrue(varN > var0)
def __setstate__(self, d):
self.__dict__.update(d)
self._tv = cm.vecD()
self._tv.extend(self.t)
self._yv = cm.vecD()
self._yv.extend(self.y)
self._dyv = cm.vecD()
self._dyv.extend(self.dy)
self._carma_process = cm.run_mcmc_carma(1, 1, self._tv, self._yv,
self._dyv, self.p, self.q, 1,
False, 1)
self._carma_process.SetMLE(True)
def setUp(self):
self.nSample = 100
self.nBurnin = 10
self.nThin = 1
self.nWalkers = 2
npts = 10
self.x = 1.0 * np.arange(npts)
ar_roots = np.array([-0.06283185-1.25663706j, -0.06283185+1.25663706j,
-0.02094395-0.25132741j, -0.02094395+0.25132741j,
-0.03141593+0.j])
sigsqr = 0.00126811439419
self.y = carmcmc.carma_process(self.x, sigsqr, ar_roots)
self.dy = np.sqrt(sigsqr) * np.ones(npts)
self.xdata = carmcmc.vecD()
self.xdata.extend(self.x)
self.ydata = carmcmc.vecD()
self.ydata.extend(self.y)
self.dydata = carmcmc.vecD()
self.dydata.extend(self.dy)
def testCar1(self):
cppSample = carmcmc.run_mcmc_car1(self.nSample, self.nBurnin, self.xdata, self.ydata, self.dydata, self.nThin)
psampler = carmcmc.Car1Sample(self.x, self.y, self.dy, cppSample)
self.assertEqual(psampler.p, 1)
psamples = np.array(cppSample.getSamples())
ploglikes = np.array(cppSample.GetLogLikes())
sample0 = carmcmc.vecD()
sample0.extend(psamples[0])
logprior0 = cppSample.getLogPrior(sample0)
loglike0 = cppSample.getLogDensity(sample0)
self.assertAlmostEqual(ploglikes[0], loglike0)
def log_likelihood(self, p):
pc = self.to_carmapack_params(p)
pcv = cm.vecD()
pcv.extend(pc)
ll = self._carma_process.getLogDensity(pcv)
if np.isnan(ll):
ll = np.NINF
return ll
def setUp(self):
self.nSample = 100
self.nBurnin = 10
self.nThin = 1
self.nWalkers = 2
npts = 10
self.x = 1.0 * np.arange(npts)
ar_roots = np.array([
-0.06283185 - 1.25663706j, -0.06283185 + 1.25663706j,
-0.02094395 - 0.25132741j, -0.02094395 + 0.25132741j,
-0.03141593 + 0.j
])
sigsqr = 0.00126811439419
self.y = carmcmc.carma_process(self.x, sigsqr, ar_roots)
self.dy = np.sqrt(sigsqr) * np.ones(npts)
self.xdata = carmcmc.vecD()
self.xdata.extend(self.x)
self.ydata = carmcmc.vecD()
self.ydata.extend(self.y)
self.dydata = carmcmc.vecD()
self.dydata.extend(self.dy)
def testCar1(self):
cppSample = carmcmc.run_mcmc_car1(self.nSample, self.nBurnin,
self.xdata, self.ydata, self.dydata,
self.nThin)
psampler = carmcmc.Car1Sample(self.x, self.y, self.dy, cppSample)
self.assertEqual(psampler.p, 1)
psamples = np.array(cppSample.getSamples())
ploglikes = np.array(cppSample.GetLogLikes())
sample0 = carmcmc.vecD()
sample0.extend(psamples[0])
logprior0 = cppSample.getLogPrior(sample0)
loglike0 = cppSample.getLogDensity(sample0)
self.assertAlmostEqual(ploglikes[0], loglike0)
def testCarpq(self, pModel=3, qModel=2):
sampler = carmcmc.run_mcmc_carma(self.nSample, self.nBurnin,
self.xdata, self.ydata, self.dydata,
pModel, qModel, self.nWalkers, False, self.nThin)
psampler = carmcmc.CarmaSample(np.array(self.xdata), np.array(self.ydata), np.array(self.dydata), sampler)
self.assertEqual(psampler.p, pModel+qModel)
psamples = np.array(sampler.getSamples())
ploglikes = np.array(sampler.GetLogLikes())
sample0 = carmcmc.vecD()
sample0.extend(psamples[0])
logprior0 = sampler.getLogPrior(sample0)
loglike0 = sampler.getLogDensity(sample0)
# OK, this is where I truly test that sampler is of class CARp and not CAR1
self.assertAlmostEqual(ploglikes[0], loglike0)
def _make_kalman_filter(self, p):
p = self.to_params(p)
ar_roots = self.ar_roots(p)
ma_coefs = self.ma_poly(p)
mu = par.bounded_values(p['logit_mu'],
low=self.mu_min,
high=self.mu_max)
nu = par.bounded_values(p['logit_nu'],
low=self.nu_min,
high=self.nu_max)
s = par.bounded_values(p['logit_sigma'],
low=self.sigma_min,
high=self.sigma_max)
sigma = s / np.sqrt(cm.carma_variance(1.0, ar_roots, ma_coefs))
sigmasq = sigma * sigma
sigmasq = float(sigmasq)
tv = cm.vecD()
tv.extend(self.t)
yv = cm.vecD()
yv.extend(self.y - mu)
dyv = cm.vecD()
dyv.extend(self.dy * nu)
arv = cm.vecC()
arv.extend(ar_roots)
mav = cm.vecD()
mav.extend(ma_coefs)
kfilter = cm.KalmanFilterp(tv, yv, dyv, sigmasq, arv, mav)
kfilter.Filter()
return kfilter
def testCarpq(self, pModel=3, qModel=2):
sampler = carmcmc.run_mcmc_carma(self.nSample, self.nBurnin,
self.xdata, self.ydata, self.dydata,
pModel, qModel, self.nWalkers, False,
self.nThin)
psampler = carmcmc.CarmaSample(np.array(self.xdata),
np.array(self.ydata),
np.array(self.dydata), sampler)
self.assertEqual(psampler.p, pModel + qModel)
psamples = np.array(sampler.getSamples())
ploglikes = np.array(sampler.GetLogLikes())
sample0 = carmcmc.vecD()
sample0.extend(psamples[0])
logprior0 = sampler.getLogPrior(sample0)
loglike0 = sampler.getLogDensity(sample0)
# OK, this is where I truly test that sampler is of class CARp and not CAR1
self.assertAlmostEqual(ploglikes[0], loglike0)
def simulate(self, p, ts, dys=None):
p = self.to_params(p)
kfilter = self._make_kalman_filter(p)
vtime = cm.vecD()
vtime.extend(ts)
ysim = np.asarray(kfilter.Simulate(vtime))
if dys is not None:
nu = par.bounded_values(p['logit_nu'],
low=self.nu_min,
high=self.nu_max)
ysim = ysim + np.random.randn(ysim.shape[0]) * dys * nu
mu = par.bounded_values(p['logit_mu'],
low=self.mu_min,
high=self.mu_max)
return ysim + mu
