# def log_pdf(theta, x, y):
# return torch.sum(pdf.log_prob(theta))
# Using log_pdf function based on MultivariateNormal torch distribution
pdf_dtype = torch.float64
pdf = MultivariateNormal(torch.zeros(2, dtype=pdf_dtype),
covariance_matrix=torch.eye(2, dtype=pdf_dtype))
def log_pdf(theta, x, y):
return pdf.log_prob(theta)
model = DistributionModel(log_pdf, 2, dtype=pdf.loc.dtype)
# %% Setup HMC sampler
sampler = HMC(model,
theta0=torch.tensor([-1, 1], dtype=model.dtype),
dataloader=DataLoader(EmptyXYDataset()),
tuner=HMCDATuner(1., e0=2.))
# %% Run HMC sampler
sampler.run(num_epochs=11000,
num_burnin_epochs=1000,
verbose=True,
verbose_step=1000)
from eeyore.models import DistributionModel
from eeyore.samplers import MALA
# %% Set up unnormalized target density
v = torch.tensor([2., 1.], dtype=torch.float64)
# def log_pdf(theta, x, y):
# return (v[0] - 1) * theta - torch.exp(theta) / v[1] + theta # Jacobian
def log_pdf(theta, x, y):
return Gamma(v[0], 1 / v[1]).log_prob(torch.exp(theta)) + theta
model = DistributionModel(log_pdf, 1, dtype=torch.float64)
# %% Setup MALA sampler
sampler = MALA(model,
theta0=torch.tensor([-1], dtype=model.dtype),
dataloader=DataLoader(EmptyXYDataset()),
step=0.25)
# %% Run MALA sampler
sampler.run(num_epochs=11000, num_burnin_epochs=1000)
# %% For convenience, name the chain list
chain_list = sampler.get_chain()