learn.probabilistic.init_log_posterior_fn

init_log_posterior_fn(likelihood, prior, train_loader, batch_size, batch_cache)

Initializes the log-posterior function.

Initializes a function that computes the log-posterior value of a parameter sample. The full log-posterior is computed batch-wise to avoid out-of-memory errors during the forward and backward pass.

Parameters:

• likelihood – Likelihood function

• prior – Prior function

• train_loader – Train data loader

• batch_size – Batch-size for batch-wise computation of the sub-likelihoods

• batch_cache – Number of batches to store in cache

Returns:

Log-posterior function