Bayesian Methods for Hackers | PyData Boston 2013

Transcript

1. Bayesian Methods \ For Hackers
2. camdp.com 10 second bio: I'm an ex-finance quant currently doing data at Cam Davidson-Pilon
3. The Future is Not Your Father's Algorithms
4. Deterministic problems: route-finding sorting database and network architectures graph and tree traversal
7. 21st Century Problems are Statistical
8. Examples of Statistical Problems These complex statistical problems: ranking search results item recommendations determining trending X forecasting ... are partially solved by modern algorithms: PageRank + personalization collaborative filtering trend detector regression models
9. Statistical Solutions (Solutions that are comfortable with randomness)
10. Bayesian Methods \ For Hackers
11. Philosophy and Interpretation of Bayesian Statistics Prior Prior + data (Prior + data) + data
12. We start with prior beliefs, which can be thought of as a summary of opinions.
13. Examples of Prior beliefs
14. Probabilities are subjective!
15. Add data... Given our prior, we can update our opinion (read: beliefs, probability), and produce a new opinion (read: new beliefs, new probability). This new distribution is called the posterior.
16. Iterate if more data is available
17. Prior + data Prior + data Prior = new Posterior = new Posterior Posterior becomes new Prior
18. Your Highschool Stats != Bayesian Stats
19. wtf?
20. Bayesian analysis relies heavily on data-visualizations, instead of confusing p-values and hypothesis testing.
21. One more important detail about Bayesian statistics
22. Bayesian statistics does not return a single number, it returns a distribution. This is both a blessing and curse: Blessing We can do very useful things with a distribution, like visualize uncertainty, significance and compute winning statistics. Curse The algorithms needed to return a distribution are intimidating.
23. Example...
24. Glorious A/B Testing (Make IPython go now)
25. Observing Dark Worlds (make IPython go now)
26. Bayesian Methods Future Developments \ For Hackers
28. 1. pip install bayesian-methods-for- hackers 2. Meta datasets 3. Adapting to the new IPython 1.0 and PyMC 3.0 4. physical copy?
29. Thanks! =)