# ziggystar

less info
reputation
29
bio website location age member for 2 years, 9 months seen 1 hour ago profile views 32

# 100 Actions

 Dec3 comment How to understand degrees of freedom? Thanks for confirming me in not understanding the deeper meaning of DF wrt. statistics. I tried to read this up many times, and it never made any sense to me. Nov29 comment Confusion: different definitions of MAP estimation in Graphical Models (MRFs) I've looked into the book by Koller and Friedman. And it appears you're right. But there are two different things (which I also mixed up). They call MAP estimation when estimating parameters; then they give your formula 1. But there is also the MAP query, asking for the most probable assignment; and this is what I thought of, and it is defined by your second formula, if you also want to have the parameters visible (which you usually don't in this case). Nov28 comment Confusion: different definitions of MAP estimation in Graphical Models (MRFs) Imho, the usual MAP problem is $\arg\max_x P(x|z)$ for some evidence $z$. This is considering the parameters of the model as fixed (parameters are usually $\theta$). This means MAP is the most probably assignment to the non-observed variables, after observing $z$. Not this is equivalent to your point 1 up to variable renaming (though it's a quite unusual renaming, considering the usual naming scheme). So I'm more with point 2., if you want to have the parameters also visible. Nov22 comment Expected value of an indicator function Since it's a random variable with two possible outcomes, it follows a Bernoulli distribution with some particular value for $p$. Calculating $p$ is not possible based only on the information given in the question. In particular we are missing what $X_i$ (probably some RV) and $h$ are (probably some constant). Nov22 comment Identifying the population and samples in a study There's also an answer of me to a similar question: stats.stackexchange.com/a/65564/3293 Nov13 revised Classifier for only one class fix itemize Nov13 suggested suggested edit on Classifier for only one class Nov7 comment Markov Decision Process and its generality Partially Observable MDPs? Yes, they can be reduced to MDPs, too. But a discrete state POMDP will be reduced to a continuous state MDP. Oct30 comment Understanding d-separation theory in causal Bayesian networks It doesn't flow from X to D, if only X is observed. You state it just below the picture. (Though you correctly describe it further down). Oct30 comment Name of phenomenon on estimated CDF plots of censored data @Glen_b Yes, it's not normalized. But does it matter? Oct30 revised Name of phenomenon on estimated CDF plots of censored data added 47 characters in body Oct28 comment How can to compare 1750 samples between 3 groups by R? What have you tried so far? Oct24 comment how many years will it take to achieve six-sigma quality? That's a homework? Oct23 comment Problem with factorial design in Minitab The free technical support or the resources to success? Oct22 comment Problem with factorial design in Minitab According to their promo video they offer free technical support and "resources to success". Oct20 revised Statistical significance in yes/no poll question changed "of" to "or" as asker stated in some comment Oct20 suggested suggested edit on Statistical significance in yes/no poll question Oct5 comment Can a Naive Bayes classificator “learn” variables which are not in the training set? How about building a joint dataset using your sets A,B,C by joining them on the variables in C? Oct1 comment Help understanding an explanation about minimum description length principle With decompression rule the author means decompression program or algorithm. Otherwise this seems to be written perfectly clear. Sep30 revised The meaning of convergence in Variational Inference? deleted 7 characters in body