from torch import optim
from ..models.model import Model
from ..utils import tolist
from ..losses import ELBO
[docs]class VI(Model):
"""
Variational Inference (Amortized inference)
"""
def __init__(self, p, approximate_dist,
other_distributions=[],
optimizer=optim.Adam,
optimizer_params={}):
# set distributions (for training)
distributions = [p, approximate_dist] + tolist(other_distributions)
# set losses
elbo = ELBO(p, approximate_dist)
loss = -elbo.mean()
super().__init__(loss, test_loss=loss,
distributions=distributions,
optimizer=optimizer, optimizer_params=optimizer_params)
[docs] def train(self, train_x={}, **kwargs):
return super().train(train_x, **kwargs)
[docs] def test(self, test_x={}, **kwargs):
return super().test(test_x, **kwargs)