MMM Open- Source Showdown- A Benchmark of PyMC-Marketing vs. Google Meridian
summary
An open benchmark compares PyMC‑Marketing and Google Meridian on ESS/s, parameter recovery, convergence diagnostics and memory use using synthetic multi‑geo ad‑spend. PyMC‑Marketing with a NumPyro backend shows ESS/s twice higher and completes the largest (≈10k×43) fit while Meridian fails to converge on largest dataset. PyMC also gives lower bias and scaled RMSSE ~2x worse for Meridian, and Meridian’s spline/adstock choices create persistent autocorrelated residuals.
AI-written summary · about
chapters (10)
- 0:00Media mix modeling frameworks
- 3:40Bayesian models in MMM
- 10:23Crash course in MMM
- 13:15Model performance at scale
- 18:57Speed vs. accuracy benchmark
- 21:55Reconstruction error and ROI
- 27:24Meridian’s seasonality limits
- 31:58Updated benchmark results
- 34:27Open-source benchmark
- 35:12Q&A
related talks
44mOpen-Source and Science in the Era of Foundation Models
14mChaos Genius: Open Sourcing Outlier Detection and Root Cause Analysis
41mOpenNMS Geschäftsbericht
11mBuilding Frontier Intelligence in the Open
29mWhy Proprietary Investment Research Platforms won't Last
02mOpen Source Force Multipliers
36mPreventing Random Acts of Metrics
24mEmbench 0.5: A Free Benchmark Suite for IoT