From 966406e2d8ced474a41ee42515590d61351e002c Mon Sep 17 00:00:00 2001
From: bsenst <benjamin.senst@gmail.com>
Date: Sat, 9 Jul 2022 17:47:16 +0200
Subject: [PATCH] Add timestamps for video 04 Aki Vehrati, Inference
 diagnostics

### Reference
Towards #11
Closes #29

### Description
Adjusted auto-timestamps and added new ones
---
 videos-list/04-aki.md | 49 +++++++++++++++++++++++++++++++++++++------
 1 file changed, 43 insertions(+), 6 deletions(-)

diff --git a/videos-list/04-aki.md b/videos-list/04-aki.md
index 5bf563d..2edd23b 100644
--- a/videos-list/04-aki.md
+++ b/videos-list/04-aki.md
@@ -13,12 +13,49 @@ Discourse Discussion
 https://discourse.pymc.io/t/keynote-these-are-a-few-of-my-favorite-inference-diagnostics-by-aki-vehtari/6180
 
 ## Timestamps
-- 0:00 Start of event
-- x:xx 
-- x:xx
-
-## Note: help us add timestamps here
-https://github.com/pymc-devs/video-timestamps
+00:00 Introduction by Aki
+00:22 Outline of the talk
+00:48 Run inference many times
+01:27 MCMC warm-up and convergence diagnostics
+03:10 It is good to run several chains
+03:49 Trace plots & convergence
+04:24 Convergence in worm plots
+04:53 Converge vs not converge
+05:10 R-hat for MCMC convergence diagnostics
+06:36 R-hat compares within and total variances - 50 warmup, 50 post warmup iterations
+08:34 Running more - 500 warmup, 500 post warmup iterations
+09:06 5000 warmup, 5000 post warmup iterations
+09:50 Total variance and within chain variance
+10:47 Overview versions of R-hat
+12:42 R-hat versions 1-4
+14:13 R-hat v1-v4 vs v5
+15:10 R-hat v5: Rank normalization and folding
+18:14 Effective sample size and Monte Carlo error
+21:55 Local effective sample size (ESS)
+24:43 Bulk-ESS and Tail-ESS
+26:50 Rank plots
+27:52 Traces vs. Rank plots
+28:22 Uniformity check?
+29:41 ECDF and ECDF difference
+32:20 ECDF difference envelope for multiple chains
+32:42 R* multivariate diagnostic
+34:45 MCMC convergence and accuracy diagnostics
+35:08 Variational inference (VI) convergence diagnostics
+37:31 Convergence diagnostic for VI optimization
+41:24 Split-R-hat
+42:59 VI accuracy diagnostics
+43:46 Importance sampling (IS)
+45:24 Importance function
+47:08 Example: normal approximation at the mode
+51:13 Effective sample size for importance sampling
+53:16 Pareto smoothed importance sampling
+54:08 ESS and MCSE for importance sampling
+54:39 Pareto k-hat diagnostic for VI
+55:36 VI convergence and accuracy diagnostics
+56:04 Stacking for non-mixing Bayesian computations
+57:48 Favorite inference diagnostics
+58:28 References
+58:38 Software references
 
 Speaker bio:
 Aki is an Associate professor in computational probabilistic modeling at Aalto University, Finland.