Rémi Louf    

@remilouf

I like Jazz, good writing, and machines that doubt themselves. On 👨‍👩‍👦‍👦🐱🐱 leave.

Paris, France
Vrijeme pridruživanja: travanj 2009.

Tweetovi

Blokirali ste korisnika/cu @remilouf

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @remilouf

  1. Going through my notes of the past few years, I think there is definitely a book to be written about Bayesian A/B testing, decision theory and causal inference. Maybe I missed it or it wouldn’t intetest anyone?

    Poništi
  2. 👇 The Engine: a framework to think about the growth of consumer app. TL;DR. 1. Map your loops 🔄 2. K-factors are meaningless 😮 3. Timescales matter more than anything else 🚀

    Poništi
  3. I committed to finish and publish everything I drafted over the past few years, so expect tweets about random topics: A/B testing, recommendations and Bayesian stuff.

    Poništi
  4. 3. velj

    What are the best practices when it comes to the first release of an open source project? When? What needs to be in place?

    Poništi
  5. 2. velj

    “The server couldn’t handle our 1,000 DAU”

    Prikaži ovu nit
    Poništi
  6. 2. velj

    “The EMR? Oh that’s because we need to parse 8Gb of text with Spark every morning”

    Prikaži ovu nit
    Poništi
  7. 2. velj

    “Yeah so we needed to move fast so we just stored the password in plain text”

    Prikaži ovu nit
    Poništi
  8. 2. velj

    “We don’t have a database backup”

    Prikaži ovu nit
    Poništi
  9. 2. velj

    Better idea 👇 Nightmare Kitchens but Linus Torvalds visiting startups.

    Prikaži ovu nit
    Poništi
  10. 1. velj

    Trying to convince my wife (producer) to pitch a show like Nightmare Kitchens but with a tech guru visiting startups.

    Poništi
  11. 31. sij
    Poništi
  12. 30. sij

    Current status: 🚀 Yeah, that’s great idea! 🎢 Oh wait there’s.... 💥 Aaaahhh it doesn’t work 📚 Ok, let’s read more 😳 Omg I don’t understand anything 😱 My life is over 💡 Oh wait, maybe... 🔁 3 times a day.

    Poništi
  13. 29. sij

    We don’t spend nearly enough time reading the doc for python’s standard library. I came for ”update_wrapper” and realized I am guilty of having reimplemented a probably buggy version of ”lru_cache” and ”register”🤦‍♂️

    Poništi
  14. proslijedio/la je Tweet
    26. sij

    "Newtonian Monte Carlo: single-site MCMC meets second-order gradient methods"

    Poništi
  15. 25. sij

    For a measured & scientifically sound analysis of the new virus’ potential spreading, I highly recommend following and his colleagues’ updates 👇

    Poništi
  16. 24. sij

    I just had a look at Murphy’s Machine Learning and it seems the term is used for different things? I meant in the post-simulation context:

    Prikaži ovu nit
    Poništi
  17. 24. sij

    Do you know of a PPL thats uses Rao-Blackwellization? Besides the extra computation time, is there a good reason NOT to do it? Do adaptive schemes make it useless in practice?

    Prikaži ovu nit
    Poništi
  18. proslijedio/la je Tweet
    24. sij

    A non-standard review of MCMC methods by Christian Robert and Wu Changye. "Non-standard" in that is augmented with questions found on Cross-validated

    Poništi
  19. 23. sij

    Whenever I read “the way you initialize the weights is very important” all I hear is “choosing a good prior is very important”

    Poništi
  20. 22. sij

    Yes theoretical physics is hard, but have you tried managing shapes in a PPL with vectorized sampling?

    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·