Bernoulli distribution

#1
I am reading the Bernoulli distribution. In the example a dado / dice with 6 faces.
By rolling 5 times a dice I have the chance to get 3 twice first, then two different numbers and finally the number ''3''?
He uses a Bernoulli distribution to calculate the composite probability. let's suppose for a moment that the random variable is not a dice (containing a finite number of faces and numbers) but is an indeterminable event on which we can only make hypotheses. The Bernoulli distribution will not provide the right results? Is not that ok?
 

hlsmith

Not a robit
#2
let's suppose for a moment that the random variable is not a dice (containing a finite number of faces and numbers) but is an indeterminable event on which we can only make hypotheses. The Bernoulli distribution will not provide the right results? Is not that ok?
Can you give an example of what you are thinking about - say like the existence of a black swan when one has never been seen?
 
#5
So you know their is a finite space, but not the number and/or the probability of each event?
In a dice every event has the same odds to happen (1/6).

I was asking how would it be the bernoulli distribution if the events would have different probabilities to happen

Let's keep the example of a black swan when no one has never seen it before. How do you trasfrom this in probability?
 
Last edited:

hlsmith

Not a robit
#6
You can use binomial distribution for categorical outcomes with different probabilities. Think of Urn problems with different number of balls. As for the Swan Dilemma that is not really my forte, but it seems like if the distribution was completely unknown, perhaps a Bayesian type of "prior" would need to come into play.
 
#7
You can use binomial distribution for categorical outcomes with different probabilities. Think of Urn problems with different number of balls. As for the Swan Dilemma that is not really my forte, but it seems like if the distribution was completely unknown, perhaps a Bayesian type of "prior" would need to come into play.
Thank you, with a Bayesian type of ''prior'' what should I assume? shoud I set as a ''prior'' probability (computed by my personal tought)?

sorry if this is a terrible question, I am new
 

hlsmith

Not a robit
#8
Ideally priors are based on prior data, such as prevalence of the event. In addition this allows you to also provide a precision measure. However, if you have no data at all to help support the prior, yeah unfortunately it comes down to your educated guess. This was something that bothered me quite a bit before I used any Bayesian analyses. Then I figured out that how you get around this subjective feeling is that you provide flat priors or large precision values. These things remove some of the weight of the prior.

Is it possible for you to just explicitly state the actual context of your question. I think that would greatly help us understand and help you.