# Advice on checking convergence with SVGD

**URL:** <https://discourse.pymc.io/t/advice-on-checking-convergence-with-svgd/3765>\
**Category:** Questions\
**Created:** [August 21, 2019, 7:39pm UTC](https://discourse.pymc.io/t/advice-on-checking-convergence-with-svgd/3765 "2019-08-21T19:39:00Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![calvinm](https://yyz2.discourse-cdn.com/flex036/user_avatar/discourse.pymc.io/calvinm/32/2129_2.png) [@calvinm](https://discourse.pymc.io/u/calvinm)\
**Post date:** [August 21, 2019, 7:39pm UTC](https://discourse.pymc.io/t/advice-on-checking-convergence-with-svgd/3765/1 "2019-08-21T19:39:00Z")

</div>

Because SVGD doesn’t optimize the ELBO, it doesn’t record the ELBO objective in hist at each iteration. I’d like to get a sense of how many iterations I need, but I don’t want to create a Tracker that saves svgd.approx.histogram.eval at every iteration, because that’s not really scalable with my high-dimensional model and rather large number of particles. Any recommendations?