unit8co/darts

Improve numerical stability of the NLL computation for some distributions

Open

#621 aperta il 1 dic 2021

Vedi su GitHub
 (0 commenti) (0 reazioni) (0 assegnatari)Python (762 fork)batch import
good first issueimprovement

Metriche repository

Star
 (6832 star)
Metriche merge PR
 (Merge medio 143g 16h) (16 PR mergiate in 30 g)

Descrizione

Not all Likelihoods have numerically-stable computaiton of the NLL. By default, we rely on the log_prob() function of the PyTorch Distribution objects, but it might not be the best way to get numerically stable gradients. I think it’s fairly easy to build cases where Negative Binomial fails for instance.

See: #663

Also for the Weibull distribution, the outputs changed when we changed the shape of the network outputs, which is weird and ideally should be investigated. Not clear if that’s due to numerical instability…

Guida contributor