7
$\begingroup$

I am trying to understand the Chinese Restaurant process (CRP) and Weighted Chinese Restaurant process (WCRP) described in a research paper "Automatic Discovery of Cognitive Skills"- Robert V. Lindsey, Mohammad Khajah, Michael C. Mozer to Improve the Prediction of Student Learning. In CRP all the implementations (cf., Infinite mixture models with nonparametric Bayes and the Dirichlet process) have a comparison made with the random number to decide if the customer chooses to sit on a new or existing table. Why this check is made and also how will this check condition differ in WCRP?

Improve this question
$\endgroup$
2
  • 2
    $\begingroup$ @MichaelChernick, what are you talking about? Of course this is related to statistics. It's a question about discrete time stochastic processes. Look at the wikipedia page about CRPs en.wikipedia.org/wiki/Chinese_restaurant_process $\endgroup$
    – GoF_Logistic
    Commented Apr 5, 2017 at 18:05
  • 2
    $\begingroup$ To get back to the question. Someone here should be able to answer the question. However, the link isn't clear on what you mean by WCRP. This isn't a standard thing, so preferably define it or link to an authoritative resource. Without that it's just guess work what the difference might be. $\endgroup$
    – conjectures
    Commented Apr 5, 2017 at 19:26

2 Answers 2

Reset to default
3
$\begingroup$

This implementation is using the Polya urn representation of the Dirichlet process like described by Blackwell and MacQueen (1973). In the link you've provided this particular part of the process is described as "With probability α/(1+α) he sits down at a new table." Conceptually one can think of this as capturing the idea that in principle there are an infinite number of possible tables to join.

The only difference under a weighted Chinese restaurant process in terms of the random number check is the probability of deciding to start a new table (cluster) will be different.

Improve this answer
$\endgroup$
0
$\begingroup$

The CRP is a model used with graphical models to simulate how many clusters you have.

It's not applied to data points. In fact, it is a prior, and does not depend on the data at all.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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