Cours 29 Bayesian stats cours
- design the model (data story)
- condition on the data (update the model)
- evaluate the model (critique)
ex : 9 times dta : W (water) or L (land)
p = proba de W proba de L = 1-p
prior : information before the data (p in [0;1]) posterior : update info of each value of p conditional on data
chaque postérior est le prior du prochain posterior plus on a de données plus il est aisé d’avoir un résulat précis
Define generative relations between the variables
W, L, W p * (1-p) *p = p2(1-p)1 : relative number to see W
Vraissemblance :
29.1 Grid approximation (to define posterior) :
posterior proba = standardizez product of proba of the data and prior proba standardisé : add up all the products and divide by this sum
grid approximation uses finite grid of parameter values instead of continuous space too expensive with more yhan a few parameters
Sampling from the posterior
Intervals : how much mass
Percentile intervals (PI): equal area in each tail Hightest posterior density intervals (HPDI) : narrowest interval containing mass
Mean nearly always more sensible than the mode
Model :