Sean J. Taylor

@seanjtaylor

I manage the Core Statistics Team at Facebook. Keywords: Experiments, Causal Inference, Statistics, Machine Learning, Economics.

San Francisco, CA
ಫೆಬ್ರವರಿ 2009 ಸಮಯದಲ್ಲಿ ಸೇರಿದ್ದಾರೆ

ಟ್ವೀಟ್‌ಗಳು

ನೀವು @seanjtaylor ಅವರನ್ನು ತಡೆಹಿಡಿದಿರುವಿರಿ

ಈ ಟ್ವೀಟ್‌ಗಳನ್ನು ವೀಕ್ಷಿಸಲು ನೀವು ಖಚಿತವಾಗಿ ಬಯಸುವಿರಾ? ಟ್ವೀಟ್ ವೀಕ್ಷಣೆಯು @seanjtaylor ಅವರ ತಡೆತೆರವುಗೊಳಿಸುವುದಿಲ್ಲ

  1. ಪಿನ್ ಮಾಡಿದ ಟ್ವೀಟ್
    ಫೆಬ್ರವರಿ 23,2017

    Prophet is an open-source forecasting package implemented in both Python and R, used in production at Facebook:

    ರದ್ದುಗೊಳಿಸು
  2. 3 ಗಂಟೆಗಳ ಹಿಂದೆ

    Great thread on some downsides of applying Bayesian methods in practice.

    ರದ್ದುಗೊಳಿಸು
  3. ಅವರು ಮರುಟ್ವೀಟಿಸಿದ್ದಾರೆ
    ಡಿಸೆಂ 19

    If you want to understand the latest NYTimes story on FB data, read this thread. Provides context and details on safeguards that are missing/glossed over in original story

    ರದ್ದುಗೊಳಿಸು
  4. ಅವರು ಮರುಟ್ವೀಟಿಸಿದ್ದಾರೆ
    ಡಿಸೆಂ 19

    Very excited to announce that I am one of 3 recipients of the Facebook Computational Social Science award this year. This grant will fund my current work in attempting to understand and improve the biases in the foster care system in Wisconsin.

    ರದ್ದುಗೊಳಿಸು
  5. ಅವರು ಮರುಟ್ವೀಟಿಸಿದ್ದಾರೆ
    ಡಿಸೆಂ 18

    What I’ve learnt this year: if you like someone’s content, TV show, podcast, blog, Twitter feed, play, whatever, tell them. You’d be surprised at how little praise people receive for doing something that’s fucking great, because you think they’re already getting it.

    ರದ್ದುಗೊಳಿಸು
  6. ಅವರು ಮರುಟ್ವೀಟಿಸಿದ್ದಾರೆ
    ಡಿಸೆಂ 18

    Bayesian optimization played a significant role for AlphaGo! "prior to the match with Lee Sedol, we tuned the latest AlphaGo agent and this improved its win-rate from 50% to 66.5% in self-play games." Very cool that DeepMind is sharing details

    ರದ್ದುಗೊಳಿಸು
  7. ಡಿಸೆಂ 18

    Then people incorrectly apply the rules, and it gets crazy quickly. I've seen people claim you shouldn't fit OLS if the covariates are correlated or aren't normally distributed or if the outcome is a count/binary variable.

    ಈ ಥ್ರೆಡ್ ತೋರಿಸಿ
    ರದ್ದುಗೊಳಿಸು
  8. ಡಿಸೆಂ 18

    "Cargo cult statistics" is widespread in practice. Much of my econometrics training amounted to a set of rules about when you were allowed to use which models/estimators. People who understand what they are doing are then arbitrarily subjected to these rules by reviewers.

    ಈ ಥ್ರೆಡ್ ತೋರಿಸಿ
    ರದ್ದುಗೊಳಿಸು
  9. ಅವರು ಮರುಟ್ವೀಟಿಸಿದ್ದಾರೆ
    ಡಿಸೆಂ 18

    This thread reminds me of a frequently recurring scene at NYU Politics circa 2010. [Argument about "proper" statistical model] Howard Rosenthal: Is the answer different if you just use OLS? Speaker: No. HR: Then why are we wasting our time arguing about this?

    ರದ್ದುಗೊಳಿಸು
  10. ಡಿಸೆಂ 17

    It is so, so hard to turn off your “I know a lot about things, so I know about this thing” instinct. I struggle with this on a regular basis. Especially common to be overconfident in things just a hop or two away from your expertise.

    ರದ್ದುಗೊಳಿಸು
  11. ಡಿಸೆಂ 17

    When you see posts by researchers misinterpreting statistical results about researchers misinterpreting statistical results. 😬

    ರದ್ದುಗೊಳಿಸು
  12. ಡಿಸೆಂ 16

    In my experience there are two important phases of empirical inquiry: 1) understanding what *might be* happening and 2) rigorously determining what *is* happening. I think we’ve collectively underinvested in (1).

    ರದ್ದುಗೊಳಿಸು
  13. ಅವರು ಮರುಟ್ವೀಟಿಸಿದ್ದಾರೆ
    ಡಿಸೆಂ 15

    Foundations of Agnostic Statistics is available for pre-order It is an accessible, lucid & rigorous introduction to key tools of statistics, econometrics & data science — while relying only on minimal assumptions.

    ಈ ಥ್ರೆಡ್ ತೋರಿಸಿ
    ರದ್ದುಗೊಳಿಸು
  14. ಅವರು ಮರುಟ್ವೀಟಿಸಿದ್ದಾರೆ
    ಡಿಸೆಂ 16

    If someone tells you writing software isn't worth it, think twice before taking that advice. That tweet was two years ago. {rio} is now installed 175,000 times per month.

    ರದ್ದುಗೊಳಿಸು
  15. ಅವರು ಮರುಟ್ವೀಟಿಸಿದ್ದಾರೆ
    ಡಿಸೆಂ 15

    New project from IBM allows you to write Stan models in pure Python! And notably the API doesn't require additional numpy, etc imports to mix in, either.

    ರದ್ದುಗೊಳಿಸು
  16. ಅವರು ಮರುಟ್ವೀಟಿಸಿದ್ದಾರೆ
    ಡಿಸೆಂ 14
    ಅವರಿಗೆ ಪ್ರತಿಕ್ರಿಯಿಸುತ್ತಿದ್ದಾರೆ

    You're right and I agree, but the trick here is that validity of the posterior also rests on the underlying inference technology, and reasoning about (and diagnosing) that can be incredibly hard.

    ರದ್ದುಗೊಳಿಸು
  17. ಅವರು ಮರುಟ್ವೀಟಿಸಿದ್ದಾರೆ
    ಡಿಸೆಂ 14

    Big news: open-sourced PyText on Github today. PyText is a PyTorch NLP framework including production ready models for various NLP/NLU tasks

    ರದ್ದುಗೊಳಿಸು
  18. ಡಿಸೆಂ 14

    Good Part 6: Interpretability of posteriors What a posterior *means* makes more intuitive sense to people than most statistical tests. Validity of posterior rests on underlying assumption about correctness of model, which is not hard to to reason about.

    ಈ ಥ್ರೆಡ್ ತೋರಿಸಿ
    ರದ್ದುಗೊಳಿಸು
  19. ಡಿಸೆಂ 14

    Good Part 5: Model checking as a core activity Good Bayesian analyses consider a wide range of models that vary in assumptions and flexibility in order to see how they affect substantive results. There are principled, practical procedures for doing this.

    ಈ ಥ್ರೆಡ್ ತೋರಿಸಿ
    ರದ್ದುಗೊಳಿಸು
  20. ಡಿಸೆಂ 14

    Good Part 4: Borrowing strength / sharing information A common feature of Bayesian analysis is leveraging multiple sources of data (from different groups, times, or geographies) to share related parameters through a prior. This can help enormously with precision.

    ಈ ಥ್ರೆಡ್ ತೋರಿಸಿ
    ರದ್ದುಗೊಳಿಸು
  21. ಡಿಸೆಂ 14

    Good Part 3: Estimating a distribution Bayesian analyses produce distributions as estimates rather than specific statistics about distributions. That means you deeply understand uncertainty and get a full-featured input into any downstream decision/calculation you need to make.

    ಈ ಥ್ರೆಡ್ ತೋರಿಸಿ
    ರದ್ದುಗೊಳಿಸು

ಲೋಡಿಂಗ್ ಸಮಯ ಸ್ವಲ್ಪ ತೆಗೆದುಕೊಳ್ಳುತ್ತಿರುವಂತೆನಿಸುತ್ತದೆ.

Twitter ಸಾಮರ್ಥ್ಯ ಮೀರಿರಬಹುದು ಅಥವಾ ಕ್ಷಣಿಕವಾದ ತೊಂದರೆಯನ್ನು ಅನುಭವಿಸುತ್ತಿರಬಹುದು. ಮತ್ತೆ ಪ್ರಯತ್ನಿಸಿ ಅಥವಾ ಹೆಚ್ಚಿನ ಮಾಹಿತಿಗೆ Twitter ಸ್ಥಿತಿಗೆ ಭೇಟಿ ನೀಡಿ.

    ಇದನ್ನೂ ಸಹ ನೀವು ಇಷ್ಟಪಡಬಹುದು

    ·