Remove iid_bridge (other PR)

Author: manuelgloeckler

sbi/inference/potentials/score_based_potential.py

@@ -230,92 +230,3 @@ def f(t, x):
     )
     return transform
 
-
-def _iid_bridge(
-    theta: Tensor,
-    xos: Tensor,
-    time: Tensor,
-    score_estimator: ConditionalScoreEstimator,
-    prior: Distribution,
-    t_max: float = 1.0,
-):
-    r"""
-    Returns the score-based potential for multiple IID observations.
-
-    This can require a special solver to obtain the correct tall posterior.
-
-    Args:
-        input: The parameter values at which to evaluate the potential.
-        condition: The observed data at which to evaluate the potential.
-        time: The diffusion time.
-        score_estimator: The neural network modelling the score.
-        prior: The prior distribution.
-    """
-
-    assert (
-        next(score_estimator.parameters()).device == xos.device
-        and xos.device == theta.device
-    ), f"""device mismatch: estimator, x, theta: \
-        {next(score_estimator.parameters()).device}, {xos.device},
-        {theta.device}."""
-
-    # Get number of observations which are left from event_shape if they exist.
-    condition_shape = score_estimator.condition_shape
-    num_obs = xos.shape[-len(condition_shape) - 1]
-
-    # Calculate likelihood in one batch.
-    # xos is of shape (num_obs, *condition_shape).
-    # theta is of shape (num_samples, *parameter_shape).
-
-    # TODO: we need to combine the batch shapes of num_obs and num_samples for both
-    # theta and xos.
-    theta_per_xo = theta.repeat(num_obs, 1)
-    xos_per_theta = xos.repeat_interleave(theta.shape[0], dim=0)
-
-    score_trial_batch = score_estimator.forward(
-        input=theta_per_xo,
-        condition=xos_per_theta,
-        time=time,
-    ).reshape(num_obs, theta.shape[0], -1)
-
-    # Sum over m observations, as in Geffner et al., equation (7).
-    score_trial_sum = score_trial_batch.sum(0)
-    prior_contribution = _get_prior_contribution(time, prior, theta, num_obs, t_max)
-
-    return score_trial_sum + prior_contribution
-
-
-def _get_prior_contribution(
-    diffusion_time: Tensor,
-    prior: Distribution,
-    theta: Tensor,
-    num_obs: int,
-    t_max: float = 1.0,
-):
-    r"""Returns the prior contribution for multiple IID observations.
-
-    Args:
-        diffusion_time: The diffusion time.
-        prior: The prior distribution.
-        theta: The parameter values at which to evaluate the prior contribution.
-        num_obs: The number of independent observations.
-    """
-    # This method can be used to add several different bridges
-    # to obtain the posterior for multiple IID observations.
-    # For now, it only implements the approach by Geffner et al.
-
-    # TODO Check if prior has the grad property else use torch autograd.
-    # For now just use autograd.
-    # Ensure theta requires gradients
-    theta.requires_grad_(True)
-
-    log_prob_theta = prior.log_prob(theta)
-
-    grad_log_prob_theta = torch.autograd.grad(
-        log_prob_theta,
-        theta,
-        grad_outputs=torch.ones_like(log_prob_theta),
-        create_graph=True,
-    )[0]
-
-    return ((1 - num_obs) * (t_max - diffusion_time)) / t_max * grad_log_prob_theta