Implementation:Pyro ppl Pyro Pyro Deterministic

Knowledge Sources	Pyro Pyro Primitives
Domains	Probabilistic_Programming, Bayesian_Inference
Last Updated	2026-02-09 00:00 GMT

Overview

Concrete tool for recording deterministic computations in Pyro model execution traces.

Description

pyro.deterministic records a named deterministic value in the execution trace. Internally it creates a sample site with a Delta distribution, but with zero entropy so it does not affect the ELBO. This makes derived quantities accessible to the Predictive class and MCMC diagnostics.

Usage

Use pyro.deterministic when you have a computed quantity (a deterministic function of latent variables) that you want to include in posterior samples. Common in MCMC workflows for tracking predicted means, transformed parameters, or summary statistics.

Code Reference

Source Location

• Repository: pyro
• File: pyro/primitives.py
• Lines: L221-246

Signature

def deterministic(
    name: str,
    value: torch.Tensor,
    event_dim: Optional[int] = None,
) -> torch.Tensor:
    """
    Record a deterministic value in the trace.

    Args:
        name: name of the deterministic site
        value: the deterministic value to record
        event_dim: optional number of rightmost event dimensions
    Returns:
        The input value unchanged
    """

Import

import pyro
# Used as: pyro.deterministic(name, value)

I/O Contract

Inputs

Name	Type	Required	Description
name	str	Yes	Unique name for the deterministic site
value	torch.Tensor	Yes	Computed value to record
event_dim	Optional[int]	No	Number of rightmost event dimensions

Outputs

Name	Type	Description
return	torch.Tensor	The same input value (pass-through)

Usage Examples

Tracking Predicted Mean

import pyro
import pyro.distributions as dist

def model(X, y=None):
    weight = pyro.sample("weight", dist.Normal(0., 1.).expand([X.shape[1]]).to_event(1))
    bias = pyro.sample("bias", dist.Normal(0., 10.))

    mean = X @ weight + bias
    pyro.deterministic("predicted_mean", mean)

    sigma = pyro.sample("sigma", dist.HalfNormal(1.))
    with pyro.plate("data", X.shape[0]):
        pyro.sample("obs", dist.Normal(mean, sigma), obs=y)

Related Pages

Implements Principle

• Principle:Pyro_ppl_Pyro_Deterministic_Computation