![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
guslChurch is a new stochastic programming language based on Scheme (Church is to Scheme what IBAL is to OCaml). The idea is that programs represent stochastic processes (they specify how the data is generated).
Statisticians always want to compute the likelihood the data under certain hypotheses (often plotted as a function of the model's parameters).
The easy way to do this is by rejection sampling, i.e. estimate the likelihood by simulating with the model and parameters, and count how often you get the same data. But this is "exponentially slow".
So these languages support smarter inference methods like MCMC. I would also like to see some support for analytic methods, possibly by integrating with algebraic packages.
Such languages tend to be based on functional languages because functional programs are easier to reason about (unlike Scheme, Church forbids mutation).
Stochastic programming languages seem like a good way to organize algorithms that work in specific situations, seeing how much they overlap, how well they generalize; making it possible to better evaluate the novelty/improvement of new approaches.
Statisticians always want to compute the likelihood the data under certain hypotheses (often plotted as a function of the model's parameters).
The easy way to do this is by rejection sampling, i.e. estimate the likelihood by simulating with the model and parameters, and count how often you get the same data. But this is "exponentially slow".
So these languages support smarter inference methods like MCMC. I would also like to see some support for analytic methods, possibly by integrating with algebraic packages.
Such languages tend to be based on functional languages because functional programs are easier to reason about (unlike Scheme, Church forbids mutation).
Stochastic programming languages seem like a good way to organize algorithms that work in specific situations, seeing how much they overlap, how well they generalize; making it possible to better evaluate the novelty/improvement of new approaches.