How to look like a statistician (...)
summary
Claim: developers can write Bayesian models without a PhD and let an engine do the heavy lifting. Uses F# prototypes and Stan, implements sampling and a brute force enumerator described as the slowest probabilistic inference engine. Applies a mixture of Gaussians to StackOverflow salaries, infers a surprising 25% mistake rate (Poland mean ≈ $24,000/yr), contrasts direct Monte Carlo, discretized enumeration and mentions MCMC and practical performance limits.
AI-written summary · about
chapters (17)
- 0:00Intro
- 0:33Looking like a statistician
- 1:29probabilistic programming
- 2:15stackoverflow developer survey
- 3:08salary distribution
- 6:05the question
- 7:57mixture distributions
- 9:32probability distributions
- 10:21sampling
- 17:20Bernoulli distribution
- 22:36Progress the comp
- 25:51Discretization
- 32:11Questions
- 32:41generative probabilities
- 33:23multinomial distribution
- 34:42probability mass function
- 36:56Gaussian distributions
related talks
36mA developer's guide to probabilistic programming
42mBayes and Bool Walk into a Bar
30mProbabilistic modelling with Scala
01h 03mAn Overview of Probabilistic Programming
43mThe Power of Probabilistic Programming
31mProbabilistic Programming: The What, Why and How
01h 07mProbabilistic Programming and Readable Models
29mProbabilistic Programming: What It Is and How It Works