Implementation of TSNPE

[michaeldeistler]

Contributions of this PR

• Implement truncation for SNPE
• Coverage diagnostic
• Tutorial on all implemented methods in the toolbox
• Documetion
• Minor fixup to one of the FAQ
• Tests for picklability

Changes to the code-base

Renamed rejection_sample_posterior_within_prior to accept_reject_sample since it is more general than to sample the posterior within the prior.

API

from sbi.inference import SNPE
from sbi.utils import get_density_thresholder, RestrictedPrior

inference = SNPE()
_ = inference.append_simulations(theta, x).train()
posterior = inference.build_posterior()

accept_reject_fn = get_density_thresholder(posterior, quantile=1e-4)
proposal = RestrictedPrior(prior, accept_reject_fn, sample_with="rejection")

from sbi.analysis import run_sbc
ranks, dap_samples = run_sbc(theta, x, reduce_fns=posterior.log_prob)

[codecov-commenter]

Codecov Report

Merging #754 (3b8ef31) into main (279809b) will decrease coverage by 0.31%.
The diff coverage is 66.30%.

@@            Coverage Diff             @@
##             main     #754      +/-   ##
==========================================
- Coverage   74.26%   73.95%   -0.32%     
==========================================
  Files          79       77       -2     
  Lines        6034     6073      +39     
==========================================
+ Hits         4481     4491      +10     
- Misses       1553     1582      +29     

Flag	Coverage Δ
unittests	73.95% <66.30%> (-0.32%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
sbi/samplers/importance/sir.py	23.07% <ø> (ø)
sbi/utils/restriction_estimator.py	76.79% <56.33%> (-6.33%)	⬇️
sbi/analysis/sbc.py	80.00% <100.00%> (+1.33%)	⬆️
sbi/inference/posteriors/direct_posterior.py	98.24% <100.00%> (+0.03%)	⬆️
sbi/samplers/rejection/rejection.py	90.24% <100.00%> (+0.24%)	⬆️
sbi/utils/__init__.py	100.00% <100.00%> (ø)
sbi/neural_nets/embedding_nets.py	60.41% <0.00%> (-39.59%)	⬇️
sbi/inference/snpe/snpe_base.py	88.81% <0.00%> (ø)
sbi/__init__.py
sbi/inference/posteriors/mcmc_posterior.py	84.83% <0.00%> (+2.80%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[janfb]

looks great! thanks a lot.

I added a couple of comments and questions. I'm wondering about the naming, e.g., replacing threshold by truncate because threshold used as a verb could be a bit unintuitive.

[janfb]

looks good!