Advanced tutorials

Our advanced tutorials provide a detailed explanation of different methods. These tutorials are meant as further reading for power-users (or users who are particularly interested in a particular feature of sbi), but they should not be required to successfully run sbi. If you are looking for brief answers to specific questions, check our how-to guide.

Prior and simulator

• Efficient handling of invalid simulation outputs

Neural nets

• Embedding nets for observations

• Customizing the density estimator

• SBI with iid data and permutation-invariant embeddings

• Vector Field Methods: FMPE and NPSE

Training

• More flexibility over the training loop and samplers

• Multi-round inference

Sampling

• Refining posterior estimates with importance sampling

Diagnostics

• Posterior Predictive Checks (PPC) in SBI

• Simulation-based Calibration in SBI

• Local Classifier Two-Sample Tests (L-C2ST)

• Model Misspecification in SBI

Visualization

• Analysing variability and compensation mechanisms with conditional distributions

• Plotting functionality

• Active subspaces for sensitivity analysis