Mat Leonard

@MatDrinksTea

ML Research Engineer, formerly teaching ML at and , Physics PhD , cat owner, tea is life, He/They

San Francisco, CA, USA
Vrijeme pridruživanja: prosinac 2008.

Tweetovi

Blokirali ste korisnika/cu @MatDrinksTea

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

  1. Prikvačeni tweet
    23. ruj 2019.

    My first full course for is online! This one is about feature engineering, how you can turn your data into useful representations for training models. Check it out and let me know what you think. I want your feedback!

    Poništi
  2. 3. velj

    ML papers that report model performance really need to start including uncertainty. An easy way to do this is once you're ready to publish, train the model ~10 times with different random seeds. Evaluate each model and report the mean and variance of the results.

    Poništi
  3. 3. velj

    Today is a good day because I woke up to a new Destroyer album:

    Poništi
  4. 31. sij

    Grew up in Kansas. Public schools gave the impression that MLK and the Civil Rights Act solved all race problems. Topeka presented Brown vs Board of Education as something to be proud of, but they were the defendants! The govt defended segregation!

    Poništi
  5. 31. sij

    Been spending the last week figuring out how to get PyTorch models running on Jetson Nanos at reasonable speeds. It's been an adventure. So far, start with small models. Convert to half-precision, then to ONNX, then to TensorRT engines and hope all the operations are supported.

    Poništi
  6. 31. sij

    Before I left I built a short course on NLP. It was just released! If you've been wanting to get started working with language data, check it out:

    Poništi
  7. proslijedio/la je Tweet
    30. sij

    Pandas 1.0 is here! * Read the release notes: * Read the blogpost reflecting on what 1.0 means to our project: * Install with conda / PyPI: Thanks to our 300+ contributors to this release.

    Poništi
  8. 29. sij

    This is exactly how I write code. Prototype in a notebook and move things that work to a module. At a certain point, the notebook becomes a test for the API. You can develop and test new features you add in the module in the notebooks.

    Poništi
  9. 29. sij

    I was walking down Valencia last night and thought "This would be so much better without cars"

    Poništi
  10. 28. sij

    I'm curious how this would work for open source projects on GitHub. You change the license to "free except for Palantir"? How would you enforce this? Could an open source dev afford the legal costs? This isn't a criticism, I support the idea, I'm literally curious.

    Poništi
  11. 27. sij

    Yessss I love Bon Appetit's videos, but I really want them to feature more of the POC staff and more non-Western cuisines.

    Poništi
  12. Poništi
  13. proslijedio/la je Tweet
    18. sij

    Ban both gov and private use of facial recognition Peter Theil-backed company scraped millions' of people's biometric data off the web (Facebook, Flickr, etc). Used it to create a face rec system allowing cops (soon others) to ID any of us, w/o suspicion

    Poništi
  14. 17. sij
    Poništi
  15. 16. sij

    This is a huge win for deploying PyTorch applications

    Poništi
  16. proslijedio/la je Tweet
    16. sij

    Theorem: If students find a statistical concept hard, the problem is the concept, not the students. Proof by example: p-values, confidence intervals and likelihood functions are "hard to understand" because they are fundamentally broken, bad ideas.

    Prikaži ovu nit
    Poništi
  17. 16. sij

    I've been using Black for code formatting recently and I've found it to be wonderful. I write a bunch of ugly code, run black on the files, and it's all magically cleaned up. Very recommended:

    Poništi
  18. 16. sij

    Today's work brought to you by Julia Jacklin

    Poništi
  19. 16. sij

    I like to imagine a group of social media managers sitting in a windowless room at Facebook, all tweeting at each other in an endless cycle

    Poništi
  20. 14. sij

    Working at a startup forces you to be really good at handling multiple projects and prioritizing them appropriately. There are hundreds of things you can be doing and if you're not careful you'll end up with decision paralysis.

    Poništi
  21. proslijedio/la je Tweet
    10. sij

    The winning entry of a Kaggle machine learning competition hacked the website to get the test labels, buried and encoded them into a "private external dataset", reloaded them during pre-processing, and added noise to not get a performance of 100%. This is pretty sad.

    Prikaži ovu nit
    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:

    ·