pomegranate: fast and flexible probabilistic modeling in python
summary
pomegranate is a Cython package that claims concrete speed and flexibility wins: mixtures, hidden Markov models, and Bayesian networks all run 2x or more faster in demos, GPU support uses cupy, and out‑of‑core plus parallel APIs use additive sufficient statistics so training needs one data pass and parallel summaries. The library actually stacks models (mixtures of Bayes nets or HMMs with mixture emissions), ships a Jupyter notebook, and shows real timing and accuracy comparisons.
AI-written summary · about
chapters (22)
- 0:00pomegranate introduction
- 1:50Probabilistic modeling
- 6:04When to use pomegranate over other tools
- 9:13Presentation goals
- 13:29Supported models
- 16:25Basic probability distributions
- 18:14Model stacking and mixtures
- 23:16Why pomegranate is fast
- 27:15Sufficient stats and parallelization
- 35:59Semi-supervised learning
- 39:54Running time comparisons with other libraries
- 43:10Gossip Girl case study
- 50:55Break and resources
- 51:17General mixture models in depth
- 57:20Hidden Markov models
- 1:06:37Bayesian networks
- 1:15:10Learning Bayesian structures
- 1:24:44Bayesian classifiers
- 1:37:42Advanced classifiers
- 1:39:51Parallel model training
- 1:40:32Documentation and community
- 1:41:16Q&A on robustness
related talks
01h 17mPomegranate: fast and flexible probabilistic models in python
30mHybrid Recommendation Systems using Probabilistic Graphical Models
01h 23mProbabilistic Graphical Models, HMMs using PGMPY
18mFast Inference for Probabilistic Graphical Models
47mThe State of the Art for Probabilistic Programming
34mFast, beautiful and easy Bayesian modelling: Can you have it all?
01h 27mProbabilistic Python: An Introduction to Bayesian Modeling with PyMC
43mProbabilistic Programming in the Real World