Source code for pixyz.models.vi

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)