New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
loosen constraints on simplexes, covariance, etc. #256
Comments
In transform.hpp, CONSTRAINT_TOLERANCE is defined as 1E-8 . How would well-written code with double-precision variables end up at single-precision? |
On 11/5/13, 8:23 PM, bgoodri wrote:
I can never tell what's a rhetorical question in e-mail, so For a start, I think our CONSTRAINT_TOLERANCE is probably (improperly) I believe the main place we're having problems is with correlation The function doing the positive-definiteness testing is here: src/stan/math/error_handling/matrix/check_pos_definite.hpp So you (Ben) probably have a better handle on the issue than the rest of us! Also, what should we print out in case of error? I don't see the point of And if somebody fixes this function, can they change this mess:
to something with the continuing operators at the start of a line to
vectorD is the diagonals of LDLT (I have no idea what that is or why it should be Can't we just replace the last two conditions with (!cholesky.isPositive())
|
Between this and the possible subtraction of two large numbers followed by a multiply, |
May need to write more of the LKJ transforms. |
Copied from #240: Jo(h)n Smith reported the following issue on the mailing list.
This is still an issue in Stan 2, where running this:
produces a stream of errors of the form:
and then finally dies with
Maybe we can (a) initialize randomly a little better by shrinking to identity matrix, or (b) turn down the stringency of our positive-definiteness tests. |
Does anyone have any idea how to fix any of this? I don't, but I'd really like someone to look at it. Ben? I'm pushing it off to 2.4.0+ so it doesn't hold us up for 2.4. |
See #817 for an issue with the Wishart that we should be able to fix. |
Requested by Dan Stowell (most recently, and ourselves in the past):
We're gettig too many errors from floating-point rounding, so we should probably relax our constraints a bit. I'm not sure exactly how to pull together a list of functions that check constraints other than reading the manual closely and pulling them out.
The text was updated successfully, but these errors were encountered: