How to resume training

In many cases, you might want to continue training after an intitial training run (but with the same data).

sbi supports this with the .train(..., resume_training=True) argument.

Main syntax

inference = NPE(prior=prior)
inference = inference.append_simulations(theta, x)
inference.train(max_num_epochs=300)  # Pick `max_num_epochs` such that it does not exceed the runtime.

with open("path/to/my/inference.pkl", "wb") as handle:
    pickle.dump(inference, handle)

# To resume training:
with open("path/to/my/inference.pkl", "rb") as handle:
    inference_from_disk = pickle.load(handle)
inference_from_disk.train(resume_training=True, max_num_epochs=600)  # Run epochs 301 until 600 (or stop early).
posterior = inference_from_disk.build_posterior()