Maro

25 ms·

1. 241.▲
   Ask HN: What's your favorite Wikipedia page?

   2 points by Maro 5y ago|1 comments
2. 242.▲
   by Maro 5y ago

   Scanning various CUPED related pages, I read that it's a way to reduce the variance, and hence p-value. But CUPED is not changing the lift value (difference [or ratio] in means) between T and C itself (or at least, not in the examples
3. 243.▲
   by Maro 5y ago

   Thanks for the pointer to CUPED! I wasn't aware of it.
4. 244.▲
   by Maro 5y ago

   Thanks for the pointer to CUPED! I wasn't aware of it. After a quick scan, it seems to me that CUPED is supposed to be run at a per-unit level (ie. per user normalization), to reduce variance, which seems to be a bit different than the
5. 245.▲
   A/B testing and the historic lift paradox(bytepawn.com)

   23 points by Maro 5y ago|16 comments
6. 246.▲
   by Maro 5y ago

   Thanks for the link!
7. 247.▲
   The Full Stack Data Scientist(bytepawn.com)

   2 points by Maro 5y ago|0 comments
8. 248.▲
   by Maro 5y ago

   You missed PaxosLease!
9. 249.▲
   by Maro 5y ago

   I believe here in Dubai Uber owns a fleet of cars, and drivers can rent/lease the vehicles. That way the vehicles are nice and standardized, eg. it's almost always a Lexus ES 350.
10. 250.▲
    by Maro 5y ago

    I'm Hungarian, born in '81, so I grew up around Trabis, Wartburgs, Skodas, Moskvitch, Ladas. My mom's first car was a Trabant. At some point in the 80s we moved to W-Germany for a few years and we took the Trabant, it was not
11. 251.▲
    Building intuition for p-values and statistical significance(bytepawn.com)

    2 points by Maro 5y ago|0 comments
12. 252.▲
    How I manage notifications to keep myself uninterrupted(bytepawn.com)

    1 points by Maro 6y ago|0 comments
13. 253.▲
    by Maro 6y ago

    You are right, but also you asked for it by being pedantic: "common" should be "come on" :)
14. 254.▲
    by Maro 6y ago

    eg. Stories, Marketplace, Live Video, Workplace, or a new type of Ad product, or it could be sth smaller. As a user on the web you perceive it as a big or small feature of FB.
15. 255.▲
    by Maro 6y ago

    Some notes from a former FB engineer (DE, E5, London). First of all, congrats to the author of the post! Even having worked there, seeing these numbers is a bit shocking :) Some comments to help cool down EU emotions in this thread. As the
16. 256.▲
    by Maro 6y ago

    I really liked Cal Newport's first book, "So Good They Can't Ignore You", published in 2012 [1]. The titular sentence is great advice, one I've been following all my life. Meanwhile a lot of time went by, I'm
17. 257.▲
    by Maro 6y ago

    I haven't read the OP, but if you're interested in this stuff, my blog is full of MC simulations, both for frequentist and Bayesian experimentation: http://bytepawn.com/tag/ab-testing.html
18. 258.▲
    by Maro 6y ago

    In general, imagine a problem space where you have millions (or much much more) of timeseries, each is potentially millions long (usually it's purely floats), and you want to perform time-series specific operations, like interpolate, e
19. 259.▲
    by Maro 6y ago

    Can you add a tldr?
20. 260.▲
    by Maro 6y ago

    I worked at FB London in 2016-2017, on the Workplace team. It was not like the OP describes. It was great, everybody was super smart and hard working, velocity was significantly higher than any other job I had (incl. my own startup). I easi
21. 261.▲
    by Maro 7y ago

    http://bytepawn.com/ Recently I wrote about A/B testing topics (lots of Monte Carlo simulations). Before that Deep Learning stuff.
22. 262.▲
    by Maro 7y ago

    Catch 22 (resisting the system), Ayn Rand books (role of self-interest, role of Elon Musks in society), Asimov's Foundation books (predicting the future with integrals), Ishmael (man vs world), Dragon's Egg (best hard scifi book,
23. 263.▲
    by Maro 7y ago

    There's various factors that contribute to underreporting: - not enough testing / 80% of people have mild/no symptoms - cases reported today were contracted ~5 days ago, so there's always roughly a doubling in the pipeli
24. 264.▲
    by Maro 7y ago

    same in Hungary
25. 265.▲
    by Maro 7y ago

    I think the reasons are: 1. In most places, the absolute numbers are still very low. If you look at it like "X in a million", X is still low, so it feels like there is no need to panic. 2. Continuing the above, if you compare X to
26. 266.▲
    by Maro 7y ago

    CMIIW: The same way age segments the CFR, so does smoking, existing conditions, etc. So if you're <40, not a smoker, no existing disease, the risk is even lower. Having said that, it's still a bad situation, since getting infec
27. 267.▲
    by Maro 7y ago

    I also saw this and found it odd. Assuming p=0.01, N=1000, k=0, the binomial probability is 0.00004317. At p=0.02, 0.03 it's even lower.
28. 268.▲
    by Maro 7y ago

    I worked on Workplace at Facebook. It's super-cheap at $1-3 per MAU. If they like it, why not pay for it. OTOH if I were FB I'd give it to CERN for free.
29. 269.▲
    by Maro 7y ago

    Depends on the idea, but there's a lot you can do before you write a single line code. 1. Of your friends/network, pick 5-10 people who could be potential customers. Ask them. 2. Think of a way to scrape emails addr of people who
30. 270.▲
    by Maro 7y ago

    I found this, but it's just ML, and not easy to read: A Survey on Compiler Autotuning using Machine Learning https://arxiv.org/abs/1801.04405