Day #1 #studyingCausality by @yudapearl, since this is a research project have to get my hands dirty & find existing code for implementing causal graphical models: DONE! library: github.com/ijmbarr/causal… code, tutorial: degeneratestate.org/posts/2018/Jul… by Iain Barr 
@yudapearl Day #1 #studyingCausality p1-6 Intro to Probability Theory 2 reasons causality analysis starts w/prob 1. uncertainty of everyday situations where type 'causes' are usually in between, not necessary, nor sufficient 2. prob theory can handle exceptions unlike common causal language
@yudapearl Day #1 #studyingCausality (1.10) P(A) is shorthand 4 P(A|K) with larger context K of background assumptions (1.11), question: technically context change (shrinkage/growth) can be handled with 1 conditional prob or nested cases might apply w/larger contexts factored in separately?
@yudapearl Day #2 #studyingCausality p6-8 Combining Predictive (prior odds) & Diagnostic (likelihood ratio) Supports, Bayes Theorem comes from combining inverse conditional probabilities, prior/posterior odds & likelihood ratios come from combining complementary forms of Bayes' theorem
@yudapearl Day #3 #studyingCausality p8-10 Random Variables and Expectations, straightforward, discrete random variables are default followed by continuous translation, another news is that I started to write my own code with causalgraphicalmodels python module but no point sharing it here
@yudapearl Day #4 #studyingCausality p11-2 Conditional Independence (X & Y) & Graphoids, P(x|y,z) = P(x|z) if P(y,z) >0 is a crucial basic building block of the whole causal calculus, is there a version that is focusing on Y being replaceable by/reducible to Z in context of X & applicable?
@yudapearl Day #5 #studyingCausality p11-2 Graphoids, tried 2 re-phrase the symmetry theorem phrased by @yudapearl as 'knowing Z, if Y tells us nothing new about X, then X tells us nothing new about Y' but couldn't, my examploid was me inviting others 2 party (Z) where X introducing me to Y
Day #5 #studyingCausality p11-2 Graphoids, a cognitive update was triggered by the weak union axiom (X⫫YW|Z)->(X⫫ Y|ZW) saying W is irrelevant information no matter whether it's part of antecedent or consequent because intuitively I interpreted consequent as relevant at first
