def guide():
mu_q = pyro.param("mu_q", Variable(self.analytic_mu_n.data + 0.094 * torch.ones(2),
requires_grad=True))
log_sig_q = pyro.param("log_sig_q", Variable(
self.analytic_log_sig_n.data - 0.11 * torch.ones(2), requires_grad=True))
sig_q = torch.exp(log_sig_q)
trivial_baseline = pyro.module("mu_baseline", pt_mu_baseline, tags="baseline")
baseline_value = trivial_baseline(ng_ones(1))
mu_latent = pyro.sample("mu_latent",
dist.Normal(mu_q, sig_q, reparameterized=False),
baseline=dict(baseline_value=baseline_value))
def obs_inner(i, _i, _x):
for k in range(n_superfluous_top + n_superfluous_bottom):
z_baseline = pyro.module("z_baseline_%d_%d" % (i, k),
pt_superfluous_baselines[3 * k + i], tags="baseline")
baseline_value = z_baseline(mu_latent.detach()).unsqueeze(-1)
mean_i = pyro.param("mean_%d_%d" % (i, k),
Variable(0.5 * torch.ones(4 - i, 1), requires_grad=True))
pyro.sample("z_%d_%d" % (i, k),
dist.Normal(mean_i, ng_ones(4 - i, 1), reparameterized=False),
baseline=dict(baseline_value=baseline_value))
def obs_outer(i, x):
pyro.map_data("map_obs_inner_%d" % i, x, lambda _i, _x:
obs_inner(i, _i, _x), batch_size=4 - i)
pyro.map_data("map_obs_outer", [self.data_tensor[0:4, :], self.data_tensor[4:7, :],
self.data_tensor[7:9, :]],
lambda i, x: obs_outer(i, x), batch_size=3)
return mu_latent
def guide():
mu_q = pyro.param(
"mu_q",
Variable(self.analytic_mu_n.data + 0.234 * torch.ones(2),
requires_grad=True))
log_sig_q = pyro.param(
"log_sig_q",
Variable(self.analytic_log_sig_n.data - 0.27 * torch.ones(2),
requires_grad=True))
sig_q = torch.exp(log_sig_q)
mu_latent = pyro.sample(
"mu_latent",
fakes.nonreparameterized_normal,
mu_q,
sig_q,
baseline=dict(use_decaying_avg_baseline=True))
def obs_outer(i, x):
pyro.map_data("map_obs_inner_%d" % i,
x,
lambda _i, _x: None,
batch_size=3)
pyro.map_data("map_obs_outer",
self.data,
lambda i, x: obs_outer(i, x),
batch_size=3)
return mu_latent
def guide():
mu_q = pyro.param(
"mu_q",
Variable(analytic_mu_n.data + torch.Tensor([-0.18, 0.23]),
requires_grad=True))
log_sig_q = pyro.param(
"log_sig_q",
Variable(analytic_log_sig_n.data - torch.Tensor([-0.18, 0.23]),
requires_grad=True))
sig_q = torch.exp(log_sig_q)
pyro.sample("mu_latent", dist.normal, mu_q, sig_q)
if map_type == "list" or map_type is None:
pyro.map_data("aaa",
data,
lambda i, x: None,
batch_size=batch_size)
elif map_type == "tensor":
tdata = torch.cat([xi.view(1, -1) for xi in data], 0)
# dummy map_data to do subsampling for observe
pyro.map_data("aaa",
tdata,
lambda i, x: None,
batch_size=batch_size)
else:
pass
def model():
mu_latent = pyro.sample("mu_latent",
fakes.nonreparameterized_normal, self.mu0,
torch.pow(self.lam0, -0.5))
def obs_inner(i, _i, _x):
for k in range(n_superfluous_top):
pyro.sample("z_%d_%d" % (i, k),
fakes.nonreparameterized_normal,
ng_zeros(4 - i, 1), ng_ones(4 - i, 1))
pyro.observe("obs_%d" % i, dist.normal, _x, mu_latent,
torch.pow(self.lam, -0.5))
for k in range(n_superfluous_top,
n_superfluous_top + n_superfluous_bottom):
pyro.sample("z_%d_%d" % (i, k),
fakes.nonreparameterized_normal,
ng_zeros(4 - i, 1), ng_ones(4 - i, 1))
def obs_outer(i, x):
pyro.map_data("map_obs_inner_%d" % i,
x,
lambda _i, _x: obs_inner(i, _i, _x),
batch_size=4 - i)
pyro.map_data("map_obs_outer", [
self.data_tensor[0:4, :], self.data_tensor[4:7, :],
self.data_tensor[7:9, :]
],
lambda i, x: obs_outer(i, x),
batch_size=3)
return mu_latent
def model():
p_latent = pyro.sample("p_latent", dist.beta, self.alpha0, self.beta0)
pyro.map_data("aaa",
self.data, lambda i, x: pyro.observe(
"obs_{}".format(i), dist.bernoulli, x, p_latent),
batch_size=self.batch_size)
return p_latent
def model():
p_latent = pyro.sample("p_latent", Beta(self.alpha0, self.beta0))
x_dist = Bernoulli(p_latent)
pyro.map_data(self.data,
lambda i, x: pyro.observe("obs", x_dist, x),
batch_size=2)
return p_latent
def map_data_vector_cuda_model(subsample_size):
mu = Variable(torch.zeros(20).cuda())
sigma = Variable(torch.ones(20).cuda())
pyro.map_data("data",
mu,
lambda i, mu: pyro.sample("x", dist.normal, mu, sigma[i]),
batch_size=subsample_size,
use_cuda=True)
def guide():
alpha_q_log = pyro.param("alpha_q_log",
Variable(self.log_alpha_n.data + 0.17, requires_grad=True))
beta_q_log = pyro.param("beta_q_log",
Variable(self.log_beta_n.data - 0.143, requires_grad=True))
alpha_q, beta_q = torch.exp(alpha_q_log), torch.exp(beta_q_log)
pyro.sample("p_latent", dist.beta, alpha_q, beta_q)
pyro.map_data("aaa", self.data, lambda i, x: None, batch_size=self.batch_size)
def map_data_iter_cuda_model(subsample_size):
mu = Variable(torch.zeros(20).cuda())
sigma = Variable(torch.ones(20).cuda())
pyro.map_data(
"data",
list(mu),
lambda i, mu: pyro.sample("x_{}".format(i), dist.normal, mu, sigma[i]),
batch_size=subsample_size,
use_cuda=True)
def model():
lambda_latent = pyro.sample("lambda_latent",
Gamma(self.alpha0, self.beta0))
x_dist = Poisson(lambda_latent)
# x0 = pyro.observe("obs0", x_dist, self.data[0])
pyro.map_data(self.data,
lambda i, x: pyro.observe("obs", x_dist, x),
batch_size=3)
return lambda_latent
def model():
mu_latent = pyro.sample("mu_latent", dist.normal,
self.mu0, torch.pow(self.lam0, -0.5))
pyro.map_data("aaa", self.data, lambda i,
x: pyro.observe(
"obs_%d" % i, dist.normal,
x, mu_latent, torch.pow(self.lam, -0.5)),
batch_size=self.batch_size)
return mu_latent
def model():
prior_dist = DiagNormal(self.mu0, torch.pow(self.lam0, -0.5))
mu_latent = pyro.sample("mu_latent", prior_dist)
x_dist = DiagNormal(mu_latent, torch.pow(self.lam, -0.5))
# x = pyro.observe("obs", x_dist, self.data)
pyro.map_data(self.data,
lambda i, x: pyro.observe("obs_%d" % i, x_dist, x),
batch_size=1)
return mu_latent
def model(means, stds):
return pyro.map_data("a", means,
lambda i, x:
pyro.map_data("a_{}".format(i), stds,
lambda j, y:
pyro.sample("x_{}{}".format(i, j),
dist.normal, x, y),
batch_size=std_batch_size),
batch_size=mean_batch_size)
def guide():
mu_q = pyro.param("mu_q", Variable(self.analytic_mu_n.data + 0.134 * torch.ones(2),
requires_grad=True))
log_sig_q = pyro.param("log_sig_q", Variable(
self.analytic_log_sig_n.data - 0.14 * torch.ones(2),
requires_grad=True))
sig_q = torch.exp(log_sig_q)
pyro.sample("mu_latent", dist.Normal(mu_q, sig_q, reparameterized=reparameterized))
pyro.map_data("aaa", self.data, lambda i, x: None,
batch_size=self.batch_size)
def model(batch_size_outer=2, batch_size_inner=2):
mu_latent = pyro.sample("mu_latent", dist.normal, ng_zeros(1), ng_ones(1))
def outer(i, x):
pyro.map_data("map_inner_%d" % i, x, lambda _i, _x:
inner(i, _i, _x), batch_size=batch_size_inner)
def inner(i, _i, _x):
pyro.sample("z_%d_%d" % (i, _i), dist.normal, mu_latent + _x, ng_ones(1))
pyro.map_data("map_outer", [[ng_ones(1)] * 2] * 2, lambda i, x:
outer(i, x), batch_size=batch_size_outer)
return mu_latent
def guide():
mu_q = pyro.param(
"mu_q",
Variable(self.analytic_mu_n.data + 0.134 * torch.ones(2),
requires_grad=True))
log_sig_q = pyro.param(
"log_sig_q",
Variable(self.analytic_log_sig_n.data - 0.09 * torch.ones(2),
requires_grad=True))
sig_q = torch.exp(log_sig_q)
q_dist = DiagNormal(mu_q, sig_q)
q_dist.reparametrized = reparametrized
pyro.sample("mu_latent", q_dist)
pyro.map_data(self.data, lambda i, x: None, batch_size=1)
def model(batch_size_outer=2, batch_size_inner=2):
mu_latent = pyro.sample("mu_latent", dist.normal, ng_zeros(1), ng_ones(1))
def outer(i, x):
pyro.map_data("map_inner_%d" % i, x, lambda _i, _x:
inner(i, _i, _x), batch_size=batch_size_inner)
def inner(i, _i, _x):
pyro.sample("z_%d_%d" % (i, _i), dist.normal, mu_latent + _x, ng_ones(1))
pyro.map_data("map_outer", [[ng_ones(1)] * 2] * 2, lambda i, x:
outer(i, x), batch_size=batch_size_outer)
return mu_latent
def model():
mu_latent = pyro.sample(
"mu_latent",
dist.Normal(self.mu0, torch.pow(self.lam0, -0.5), reparameterized=False))
def obs_outer(i, x):
pyro.map_data("map_obs_inner_%d" % i, x, lambda _i, _x:
obs_inner(i, _i, _x), batch_size=3)
def obs_inner(i, _i, _x):
pyro.observe("obs_%d_%d" % (i, _i), dist.normal, _x, mu_latent,
torch.pow(self.lam, -0.5))
pyro.map_data("map_obs_outer", self.data, lambda i, x:
obs_outer(i, x), batch_size=3)
return mu_latent
def map_data_iter_model(subsample_size):
mu = Variable(torch.zeros(20))
sigma = Variable(torch.ones(20))
def local_model(i, unused):
pyro.sample("x_{}".format(i), dist.normal, mu[i], sigma[i])
return i
return pyro.map_data('mapdata', range(20), local_model, batch_size=subsample_size)
def guide():
mu_q = pyro.param("mu_q", Variable(self.analytic_mu_n.data + 0.234 * torch.ones(2),
requires_grad=True))
log_sig_q = pyro.param("log_sig_q", Variable(
self.analytic_log_sig_n.data - 0.27 * torch.ones(2),
requires_grad=True))
sig_q = torch.exp(log_sig_q)
mu_latent = pyro.sample(
"mu_latent",
dist.Normal(mu_q, sig_q, reparameterized=False),
baseline=dict(use_decaying_avg_baseline=True))
def obs_outer(i, x):
pyro.map_data("map_obs_inner_%d" % i, x, lambda _i, _x:
None, batch_size=3)
pyro.map_data("map_obs_outer", self.data, lambda i, x:
obs_outer(i, x), batch_size=3)
return mu_latent
def model():
mu_latent = pyro.sample("mu_latent", dist.normal,
mu0, torch.pow(lam0, -0.5))
if map_type == "list":
pyro.map_data("aaa", data, lambda i,
x: pyro.observe(
"obs_%d" % i, dist.normal,
x, mu_latent, torch.pow(lam, -0.5)), batch_size=batch_size)
elif map_type == "tensor":
tdata = torch.cat([xi.view(1, -1) for xi in data], 0)
pyro.map_data("aaa", tdata,
# XXX get batch size args to dist right
lambda i, x: pyro.observe("obs", dist.normal, x, mu_latent,
torch.pow(lam, -0.5)),
batch_size=batch_size)
else:
for i, x in enumerate(data):
pyro.observe('obs_%d' % i,
dist.normal, x, mu_latent, torch.pow(lam, -0.5))
return mu_latent
def map_data_vector_model(subsample_size):
mu = Variable(torch.zeros(20))
sigma = Variable(torch.ones(20))
def local_model(batch, unused):
pyro.sample("x", dist.normal, mu[batch], sigma[batch])
return batch
LongTensor = torch.cuda.LongTensor if torch.Tensor.is_cuda else torch.LongTensor
ind = Variable(LongTensor(range(20)))
batch = pyro.map_data('mapdata', ind, local_model, batch_size=subsample_size)
return list(batch.data)
def model():
mu_latent = pyro.sample(
"mu_latent",
dist.Normal(self.mu0, torch.pow(self.lam0, -0.5), reparameterized=False))
def obs_inner(i, _i, _x):
for k in range(n_superfluous_top):
pyro.sample("z_%d_%d" % (i, k),
dist.Normal(ng_zeros(4 - i, 1), ng_ones(4 - i, 1), reparameterized=False))
pyro.observe("obs_%d" % i, dist.normal, _x, mu_latent, torch.pow(self.lam, -0.5))
for k in range(n_superfluous_top, n_superfluous_top + n_superfluous_bottom):
pyro.sample("z_%d_%d" % (i, k),
dist.Normal(ng_zeros(4 - i, 1), ng_ones(4 - i, 1), reparameterized=False))
def obs_outer(i, x):
pyro.map_data("map_obs_inner_%d" % i, x, lambda _i, _x:
obs_inner(i, _i, _x), batch_size=4 - i)
pyro.map_data("map_obs_outer", [self.data_tensor[0:4, :], self.data_tensor[4:7, :],
self.data_tensor[7:9, :]],
lambda i, x: obs_outer(i, x), batch_size=3)
return mu_latent
def outer(i, x):
pyro.map_data("map_inner_%d" % i, x, lambda _i, _x:
inner(i, _i, _x), batch_size=batch_size_inner)
def model():
lambda_latent = pyro.sample("lambda_latent", dist.gamma, self.alpha0, self.beta0)
pyro.map_data("aaa",
self.data, lambda i, x: pyro.observe(
"obs_{}".format(i), dist.poisson, x, lambda_latent), batch_size=3)
return lambda_latent
def outer(i, x):
pyro.map_data("map_inner_%d" % i,
x,
lambda _i, _x: inner(i, _i, _x),
batch_size=batch_size_inner)
def obs_outer(i, x):
pyro.map_data("map_obs_inner_%d" % i, x, lambda _i, _x:
None, batch_size=3)
def model():
return pyro.map_data("md",
data,
local_model,
batch_size=1,
batch_dim=batch_dim)
def obs_outer(i, x):
pyro.map_data("map_obs_inner_%d" % i,
x,
lambda _i, _x: obs_inner(i, _i, _x),
batch_size=4 - i)
def obs_outer(i, x):
pyro.map_data("map_obs_inner_%d" % i,
x,
lambda _i, _x: None,
batch_size=3)
def obs_outer(i, x):
pyro.map_data("map_obs_inner_%d" % i, x, lambda _i, _x:
obs_inner(i, _i, _x), batch_size=4 - i)
