# Technical Working Group

**URL:** <https://discourse.mc-stan.org/t/technical-working-group/6758>\
**Category:** Developers\
**Created:** [November 28, 2018, 12:42am UTC](https://discourse.mc-stan.org/t/technical-working-group/6758 "2018-11-28T00:42:47Z")\
**Posts on this page:** 1\
**Showing post:** 7

<div class="post-metadata">

**Author:** ![Bob\_Carpenter](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.mc-stan.org/bob_carpenter/32/9230_2.png) [@Bob\_Carpenter](https://discourse.mc-stan.org/u/Bob_Carpenter)\
**Post date:** [November 30, 2018, 1:20am UTC](https://discourse.mc-stan.org/t/technical-working-group/6758/7 "2018-11-30T01:20:25Z")

</div>

Here’s a relevant recent discussion around VI and diagnostics:

> [@Adjusting default tol\_rel\_obj threshold in vb](https://discourse.mc-stan.org/t/adjusting-default-tol-rel-obj-threshold-in-vb/6745):
>
> The recent paper by Yao et al. on Stan’s [variational inference](https://arxiv.org/pdf/1802.02538.pdf) is pretty clear that lowering the threshold for change in ELBO will improve the fit of the approximated posterior and is more likely to pass the diagnostic checks they propose (see figure 2 in the paper). However, the current default in rstan, option tol\_rel\_obj, is still set at 0.01. Given the fact that rstan settings for full Bayesian inference are conservative (2000 iterations), it would seem a good practice to change the default …

It’s about adding K-hat diagnostics, changing thresholds on VI runs, etc. I’m not sure it requires a committee meeting to sort out.

---

_[View the full topic](https://discourse.mc-stan.org/t/technical-working-group/6758)._