When there are many sets of correlated samples, we basically have an empirical joint distribution or a joint sample (different names, same thing here). Once you have this, there are some nice things you could do with it.
You can answer questions like:
"Conditional on b < 20, what is the distribution for a?"
"If b = 30, what is the distribution for what a is?"
"What combinations of params lead to the highest value of a?"
Which sets of distributions are correlated with each other? By how much?
... etc.
It would be neat to explore design decisions here.
This can be useful for functions / function modules, when you're dealing with a cluster of some highly correlated variables.
I imagine there are also some really neat UIs to show this information.
Description of suggestion or shortcoming:
When there are many sets of correlated samples, we basically have an empirical joint distribution or a joint sample (different names, same thing here). Once you have this, there are some nice things you could do with it.
You can answer questions like:
b < 20
, what is the distribution fora
?"b = 30
, what is the distribution for whata
is?"a
?"It would be neat to explore design decisions here.
This can be useful for functions / function modules, when you're dealing with a cluster of some highly correlated variables.
I imagine there are also some really neat UIs to show this information.