0
$\begingroup$

Suppose I have $M$ samples from unknown distributions $F(X)$ and $F(Z|X)$. Is there a way from these two vectors to get samples of $F(X|Z)$? I understand Bayes rule, but I only know how to apply it to probability density and distribution functions, not random samples from these distributions.

$\endgroup$
2
$\begingroup$

One way is to estimate $F(X)$ and $F(Z \mid X)$ from the samples, use Bayes rule rule to obtain $F(X \mid Z)$, then sample from it.

Or, if you have many samples that are pairs of values $(x, z)$ drawn from the joint distribution $F(X, Z)$, you could 'slice' these to obtain subsamples of $F(X \mid Z)$. For example, if $Z$ is discrete and you're interested in the case $Z=z$, select out all your sampled $X$ values where the corresponding $Z=z$. If $Z$ is continuous, you could define a threshold $\epsilon$, then select out the sampled $X$ values with corresponding $Z$ values $\in [z-\epsilon, z+\epsilon]$. In this case, there's a tradeoff between the accuracy and the number of samples you get.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.