Bayesian Methods for Hackers - Chapter 2
Key: 🔑 key point 👍 minor point 📗 vocab word 📸 flashcard ❓ Question ⚙️ API
❓ This PyMC interface is odd and not very Pythonic. Why assign all variables in a context to a model instance? Lolz.
📸 What's the difference between deterministic and stochastic variables?;;Stochastic: even if all inputs and parameters were known, output would still be random (most distributions).
Deterministic: when inputs/params are known, so is output. Stochastic var + 1, for example.;
👍 Why bother marking deterministic functions as pm.deterministic? It doesn't change their behavior.;;Because it makes the variable "tracked by our sampling".
👍 PyMC : Theano :: Pyro : PyTorch
⚙️ theano.tensor.stack: combine separate Theano variables into one variable ⚙️ pm.Categorical: stochastic, takes a Theano variable
🔑 How should you approach Bayesian modeling? Imagine how you would generate the dataset - choose random variables with the right properties. For parameters, choose priors that either reflect total uncertainty (e.g., a uniform distribution), or skip setting them entirely (like alpha params for Poisson dsi
Poisson random variable vs poisson distribution?