hesenp

N = 1000
for name, quantiles in guide.quantiles(torch.arange(0., N) / N).items():
  quantiles = np.array(quantiles)
  pdf = 1 / (quantiles[1:] - quantiles[:-1]) / N
  x = (quantiles[1:] + quantiles[:-1]) / 2
  sns.plt.plot(x, pdf, label=name)
  
sns.plt.legend()
sns.plt.ylabel('density')

hesenp

pyro.clear_param_store()
  
adam_params = {"lr": 0.01, "betas": (0.90, 0.999)}
optimizer = optim.Adam(adam_params)

svi = infer.SVI(model, 
                guide, 
                optimizer, 
                loss=infer.Trace_ELBO())

hesenp

def model(x, y, truncation_label):
  a_model = pyro.sample("a_model", dist.Normal(0, 10)) 
  b_model = pyro.sample("b_model", dist.Normal(0, 10))
  
  link = torch.nn.functional.softplus(a_model * x + b_model) 
  
  with pyro.plate("data"):
    y_hidden_dist = dist.Exponential(1 / link) 
    
    with pyro.poutine.mask(mask = (truncation_label == 0)): 

hesenp

pyro.clear_param_store()

# note [1] 
hmc_kernel = HMC(model,
                 step_size = 0.1, 
                 num_steps = 4)


# Note [2] 
mcmc_run = MCMC(hmc_kernel, 

hesenp

def model(x, y, truncation_label): ## Note [1] 
  a_model = pyro.sample("a_model", dist.Normal(0, 10)) ## Note [2] 
  b_model = pyro.sample("b_model", dist.Normal(0, 10))
  
  link = torch.nn.functional.softplus(a_model * x + b_model) ## Note [3] 
  
  for i in range(len(x)):
    y_hidden_dist = dist.Exponential(1 / link[i]) ## Note [4] 
    
    if truncation_label[i] == 0: 

hesenp

n = 500
a = 2
b = 4
c = 8

x = dist.Normal(0, 0.34).sample((n,)) # Note [1]
  
link = torch.nn.functional.softplus(torch.tensor(a*x + b))
# note below, param is rate, not mean
y = dist.Exponential(rate=1 / link).sample()  

hesenp

import pyro 
import torch 
import seaborn as sns 
import pyro.distributions as dist 
from pyro import infer, optim
from pyro.infer.mcmc import HMC, MCMC
from pyro.infer import EmpiricalMarginal

assert pyro.__version__.startswith('0.3')