nmca

7 ms·

361.
by nmca 6y ago
I wanted this at university and tried to write something similar before inevitable employment and distraction. I wish you every success, a couple of ideas that may be applicable: Probably quite cheap to pay people to onboard sets of well es
362.
by nmca 6y ago
Does Chris Lattner still work at Google?
363.
by nmca 6y ago
You can run a k steps of adam-sgd and differentiate through the learning rate and scalarising parameters of a composite loss function in order to meta-learn them. Support for general n-th total derivatives is rather good :)
364.
by nmca 6y ago
JAX supports this; as do other performance-targeting numerical libraries. There are hoops of course.
365.
by nmca 7y ago
Thank you, in parallel to this response I looked at the paper in more detail and updated the original comment. A shame, I was briefly optimistic about the news.
366.
by nmca 7y ago
Ah, you imply that the underreporting rate fell over time and the headline number is from before the travel ban.
367.
by nmca 7y ago
Could you elaborate?
368.
by nmca 7y ago
I have been considering other ways to estimate this, and looked at numbers for Singapore, which I believe has particularly strict testing. They show a much lower mortality rate than the other estimates I have heard, but only at n=250. This
369.
by nmca 7y ago
Does this mean the mortality rate is, approximately, 10x less than currently estimated? (86% of cases unreported, eg about 1 in 10 reported, assume that number of unreported fatal cases is 0) Edit: closer reading suggests the optimism is sa
370.
by nmca 7y ago
What do you think will happen longer term in South Korea? (Not meant to be provocative, curious about your forecast)
371.
by nmca 7y ago
That is indeed a good summary, thanks!
372.
by nmca 7y ago
Brit here: I'm not a fan of Dominic's government, but the response seems to be well motivated by a series of reasonable arguments. As I understand it: 1. We either get herd immunity via infection or cure/end every case of the
373.
by nmca 7y ago
irrc 50k at the rate of his salary until he got a job, may be mis-remebering.
374.
by nmca 7y ago
It's actually 50k of time off (not made clear other than in the contract at the bottom), which is not exactly the same thing.
375.
by nmca 7y ago
jumpira.io : a new Jira integration that provides jump scares and horrific imagery to motivate your subordinates into writing better login pages!
376.
by nmca 7y ago
"obviously" you should do this and the rank results under a language model (mainly joking) edit: oooh and use beam search to limit the output size/time
377.
by nmca 7y ago
Very, very slow for me on the first loading screen on pixel 3 XL, mobile 4g, both chrome and mobile Firefox. Firefox loaded after 4s or so, chrome did not after 10s
378.
by nmca 7y ago
Did you read the article? Irrc you can lerp rotors and the results are smooth and interpretable...
379.
by nmca 7y ago
Indeed a germline modification to produce only male offspring would do it. Children of men, and all that.
380.
by nmca 7y ago
Seems superficially similar to roam research, which has been good to me so far.
381.
by nmca 7y ago
On the flip side, I learned low-level programming in rust, and am thankful that $bigcorp uses modern c++ so the initial gap was smaller. Rust is still better though.
382.
by nmca 7y ago
Not all charities are like this; the against malaria foundation for example seems very cost effective
383.
by nmca 7y ago
It's just a term of art, see also supervised learning, etc etc
384.
by nmca 7y ago
^^^^^ This is extremely insightful, and now ranks as the most accurate and succinct explanation I've seen of why point-free is terrible.
385.
by nmca 7y ago
It's easy to criticize the bank because they're clearly in the wrong here. Names and addresses are provided and don't match at all
386.
by nmca 7y ago
That seems a fairly specific meaning of the term that may not be in wider use, see eg: https://arxiv.org/abs/1803.10122
387.
by nmca 7y ago
There are certainly challenges, but "incorporating a world model" has been going well recently: "Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model" https://arxiv.org/abs/1911.0826
388.
by nmca 7y ago
Clips your data, changing the behaviour of smoothers etc, iirc gg.cartesian(ylim=...) is the thing that people usually want.
389.
by nmca 7y ago
Oh, seaborn is tightly coupled to matplotlib, so I don't really distinguish the two in my complaining, but you should definitely be using seaborn.
390.
by nmca 7y ago
Everyone I know at $big_research_lab uses it and hates it. But what are the alternatives really? Altair plots are larger than you dataset, plotnine has totally unusable docs and insane behaviour in some places (look up what plotnine'

More ›