I have an issue computing the integral of a function in rstan. With “integrate_1d”, the integral cannot be computed over an interval, whereby it throws up “error estimate of integral above zero exceeds the given relative tolerance times norm of integral above zero” . However, if I do the integrati…

Thanks for reporting the error and the code! So this: Exception: integrate: error estimate of integral above zero 9.59331e-10 exceeds the given relative tolerance times norm of integral above zero (in 'unknown file name' at line 150) turned out to be a problem with how the tolerances were working…

Yeah, integrate_1d can be rather frustrating. The integrate_1d function is basically a wrapper around the Boost 1d quadrature stuff over here: https://www.boost.org/doc/libs/1_74_0/libs/math/doc/html/quadrature.html . On top of doing the integrals, it will also compute gradients of these integrals…

Hi, thanks, I checked already the pdf and cdf and they are fine in the domain where the integral fails to evaluate. I noticed that “integrate” does not rely on the same algorithm for integration as “integrate_1d”. When looking into the code for “integrate”, it call on “C_call_dqags”, which apparent…

Do you have any luck turning down the tolerance for integrate_1d?

Ugh, that’s no good. What is the value of the function at one of the spots where the integration fails? (and the error message at that spot too – I’m wondering how close the integrator got) Can you use the ODE integrator to do the integral? Can you post a plot of rank_density for a set of parame…

The code I already posted will generate the error messages. For reference, I get: > order(-10,0.67,theta = c(t,A,v1,v2,s), x_r = double()) Error in order(-10, 0.67, theta = c(t, A, v1, v2, s), x_r = double()) : Exception: integrate: error estimate of integral above zero 9.59331e-10 exceeds the g…

Thanks for the code. I’ll have a look. [image] alescia: I do not know how to use the ODE integrator for such things, isn’t it meant for ordinary differential equations? The ODE integrator computes y(t) given y' = f(t, y) and y(t0). If your f is not a function of y, then the solution of the …

I played around with the code a bit. It looks to me like it’s able to do your integral [-10, 1] but then it won’t do your integral on [-10, 0.67]. Yeah that seems crazy to me. Is that a fair summary of the problem?

Yes, that is why I am asking for pointers why this is so here!

I loosened up the tolerances a bit and the function seems to work for me now: v = integrate_1d(rank_density, down, up, theta, x_r, x_i, 1e-6); I’m not sure how well it’s working though. I’m wondering if there isn’t something screwed up in how we’re interpreting the error checks from Boost. Here’s…

Difference in behavior between integrate_1d and integrate

Modeling

bbbales2 January 27, 2021, 12:41pm 17

Release is out that addresses this: Release of CmdStan 2.26.0

Topic		Replies	Views
Stabilising survival model integrand for use in integrate_1d Modeling techniques , fitting-issues	20	1575	February 1, 2021
Calling integrate_1d() reports parsing error (PARSER EXPECTED ")") Modeling integrator	24	1590	December 9, 2020
Problem on how to use Integrate_1d function Developers	17	1430	July 26, 2021
Problem using integrate Modeling	2	384	November 9, 2020
Integrate_1d not convergent in for loop Modeling rstan	4	668	October 20, 2020

Difference in behavior between integrate_1d and integrate

Related topics