Скачать или смотреть "Tensorflow Probability" by Melinda Thielbar - Research Triangle Analysts

Tensorflow Probability: What it is, and Why you'll love it! with Dr. Melinda Thielbar

RTA's first Zoom meeting went off without a hitch. We'll probably be doing it again in May.

Tensorflow probability is a new package from Google that enables probabilistic programming. In its simplest form, it allows you to specify the distribution of the target variable and estimate its parameters in just a few lines of code. Instead of shoe-horning a known function of X and Y into one of several pre-programmed and varyingly inappropriate functional forms, Tensorflow Probability is flexible enough to accommodate the most appropriate model for the problem at hand and powerful enough to estimate the parameters to the best of your data’s ability.

Is your outcome a count? Maybe something like the number of new cases of a novel disease that appear every day? Use a Poisson as the statistics gods intended.
Are you trying to estimate far too many effects? Use a Bayesian model and let your priors handle the messy middle.
Do you have too much data for your laptop’s tiny memory? Too much processing for any one CPU to finish in a year? Tensorflow Probability is GPU-enabled. Give your little CPU a rest and actually see the results of your labors while you’re still interested enough to enjoy them.

Copy of the notebook used in the talk: https://github.com/BowTieMLP/talks_an...