Stack Exchange
log in

njuffa

 Discussion on question by Mike: How d…

Imported from a comment discussion on math.stackexchange.com/q...
njuffa
Dec 12, 2024 09:32
@Adam Rubinson The question inquires about this gentleman: en.wikipedia.org/wiki/B%C3%A9la_Sz%C5%91kefalvi-Nagy : "Szőkefalvi-Nagy collaborated with Alfréd Haar and Frigyes Riesz, founders of the Szegedian school of mathematics. He contributed to the theory of Fourier series and approximation theory. His most important achievements were made in functional analysis, especially, in the theory of Hilbert space operators."
 

 Discussion on answer by littleadv: Is…

Imported from a comment discussion on travel.stackexchange.com...
njuffa
Feb 22, 2024 19:10
@littleadv Who he? FWIW, the question does not mention "Tucker Carlson". I know ALDI from Germany in the 1980s, and chained shopping carts released by inserting a coin were not a thing back then. But by the time I visited Germany in the mid 1990s ALDI shopping carts had acquired the coin locks still in use today. I did not observe any homeless people pushing shopping carts there or abandoned shopping carts in various parts of town like I see in the Bay Area. Locking the carts was all about driving down cost, something ALDI excels at.
njuffa
Feb 22, 2024 19:10
@littleadv Re "This is much more effective than a ... deposit against carts being stolen." As far as I understand, the motivation in the case of ALDI was not primarily theft prevention, but saving on labor to collect carts from the parking lot. ALDI is famous for thriving on a 1% profit margin, and numerous measures to keep costs low: In the olden days their cashiers had to memorize the price of every item sold in their stores (~ 1000 SKUs): no need for expensive scanners. Early stores did not even have a phone: store managers passed list of deliverables for next day to delivery drivers.
 

 Discussion on answer by Dan Romik: In…

Imported from a comment discussion on academia.stackexchange.c...
njuffa
Feb 20, 2024 03:41
@TimRias Could it be simply a difference in cultural expectations? I recall complaints about rudeness from Europeans when they heard nothing further after interviewing with various Silicon Valley computer industry companies. I explained to them that this is not considered rude but perfectly normal in this industry in this geographic location. Only if the company is interested in pursuing further steps will an applicant be contacted again.
 

 Discussion on answer by T.E.D.: Did e…

Imported from a comment discussion on history.stackexchange.co...
njuffa
Sep 11, 2023 13:04
Anecdotally, from examples across numerous episodes of "Finding Your Roots" with Henry Louis Gates, both of your findings appear to be on the mark. (+1).
 

 Discussion on question by mina nageh:…

Imported from a comment discussion on history.stackexchange.co...
njuffa
Apr 23, 2023 22:56
After reading Felicity Harley, "Crucifixion in Roman Antiquity: The State of the Field", Journal of Early Christian Studies, Vol. 27, No. 2, Summer 2019, pp. 303-323 and Matthew W Maslen and Piers D Mitchell, "Medical theories on the cause of death in crucifixion", J R Soc Med 2006;**99**:185–188, I conclude that we know next to nothing with certainty about details of crucifixion. Latter paper contemplates possibility of reuse: "The tip of the nail was bent, suggesting that during its insertion it had perhaps met a hard knot of wood or pre-existing nail left from an earlier crucifixion."
 

 Discussion on answer by Ottie: Why do…

Imported from a comment discussion on academia.stackexchange.c...
njuffa
Dec 9, 2022 15:41
(+1) Anecdotal corroboration: I went to the US for a 1-year student exchange after completing 10th grade in a Central European country. The American high school (in a blue collar neighborhood at the periphery of a large city in the Midwest) assigned me to the 12th grade. I graduated. I returned to my native country and was assigned to the 11th grade. Stratification occurred at age 11, I was on a STEM path bound for university. As for math: derivatives incl. chain rule were covered in the 10th grade; integrals, etc in 11th grade. Mandatory, not optional. I learned analytic geometry after that.
 

 Discussion between imallett and njuffa

Imported from a comment discussion on math.stackexchange.com/q...
njuffa
Oct 27, 2022 08:30
I meant to recommend the use of FMA to you, but I see you already use it. I haven't really designed anything without FMA since about 2008, other than for a few SO/SE questions. I do a lot of optimizing of polynomial minimax approximations and have good intuition what simulated annealing parameters to use for that. I never do as well when optimizing anything that is not a straightforward minimax polynomial; so maybe it is simply lack of experience in choosing good parameters for other use cases.
njuffa
Oct 27, 2022 07:00
Your optimizer seems to work a lot better than mine. What is it based on? The best results reported by mine so far seem to point in the same direction as your results, but are clearly not as accurate: `optimizing for rel. error: maxrelerr = 3.76984996e-4 {0x44c3576a, 0x1.ffb84ep-1, 0x1.621ca6p-4, 0x1.887eb4p-5}
optimizing for abs. error: maxabserr = 1.99183959e-4 {0x44c1c320, 0x1.fd0a36p-1, 0x1.5706a0p-4, 0x1.109f78p-4}`
njuffa
Oct 26, 2022 20:48
That seems very plausible. I am running some experiments with my own heuristic optimizer right now (related to simulated annealing) to see whether I can get results that match yours. I dialed in a very high initial "temperature", and very slow "cooling" so as to skip local minima well, but -- unsurprisingly -- this causes very slow convergence.
njuffa
Oct 26, 2022 07:35
OK, I think I got it now. You were quoting maximum absolute error on (0.0031308,1] for pow_5over12 based on changes to root_m12. I get maxabserr=1.80187870e-004 @ 0x1.5e001ap-1 for that. arg= 0x1.5e001ap-1 res=8.53246212e-001 0x1.b4dcb0p-1 ref=8.5342639987547209e-001 0x1.b4f44e1a042f0p-1
njuffa
Oct 26, 2022 07:02
Still no luck. I get maxabserr=1.20581390e-3 @ 0x1.67fff8p-6. arg= 0x1.67fff8p-6 res= 0x1.6034d6p+0 ref= 0x1.5fe5cfccb6d8ap+0.
njuffa
Oct 26, 2022 06:48
Well, your comment said "maximum error", not "maximum absolute error". Absoloute versus relative error will definitely make a difference. Let me add that to my stats.
njuffa
Oct 26, 2022 06:44
@imallett I am having trouble reproducing your result. On [2**(-126), 1], I see maximum relative error of $8.7741\times10^{-4}$ @ 0x1.69ffe6p-6 with this code: float root_m12 (float a) { float r, h; r = uint32_as_float (magic - (float_as_uint32 (a) >> 16) * 0x1555); h = 0x1.ff078cp-1f - pow_12 (r) * a; r = (0x1.61bf5ap-4f + 0x1.905614p-5f * h) * h * r + r; return r; }. I used icc with /fp:strict to compile.
njuffa
Oct 26, 2022 06:17
The simple way here is to optimize initial approximations for rootn() once, standalone fashion. Then append whatever iteration one desires, straight from the math book. If the accuracy is sufficient for the stated purpose, one is done. Optimal coefficients make sense when maximum accuracy is required, generally requires a heuristic search, and coefficients are likely to change even based on how the iteration is implemented, e.g. with FMA or without FMA. I do that for math libraries.
njuffa
Oct 26, 2022 05:56
@imallett Optimizing the "magic" constants in the initial approximations in the specific context of a subsequent iteration is definitely the way to go where maximum accuracy for the iteration is desired; it did not seem necessary for the sRGB conversion here. As for Halley vs Householder: I am a CS guy and not a mathematician; in my understanding Householder iterations are a generalization of the principles employed in Newton's second order and Halley's third order scheme (and conversely, Newton and Halley then can be viewed as specializations of the general Householder iteration).
njuffa
Oct 26, 2022 05:56
@imallett Jörg Arndt, "Matters Computational: Ideas, Algorithms, Source Code", Springer 2011, section 29.3: $d^{-1/a} = x(1-y)^{-1/a}$ where $y := (1-x^{a} d)$. Expanding as a series in $y$: $d^{-1/a} = x \frac{1}{\sqrt[a]{1-y}} = x \left[1+\frac{y}{a} + \frac{(1+a)y^{2}}{2a^{2}} + \frac{(1+a)(1+2a)y^{3}}{6a^{3}} + \ldots \right]$.
njuffa
Oct 26, 2022 05:56
@imallett I do not recall where I originally learned about this, maybe this: High order algorithms to find square roots and inverses, section 4.
 

 Discussion on answer by IMSoP: Did Ru…

Imported from a comment discussion on skeptics.stackexchange.c...
njuffa
Jan 15, 2022 18:36
@Narusan Dr. Johann Steiner published under exactly that name. In his book "Compendium der Kinderkrankheiten", published in Leipzig in 1872 (for a later English edition, see my previous comment), he writes on pp. 390-391 that based on his research, smallpox (variola) and chickenpox (varicella) do appear to be different diseases, and that he has managed to transmit varicella between patients using fluid from varicella blisters, with an observed incubation time of eight days.
njuffa
Jan 15, 2022 18:36
Johann Steiner, Compendium of children's diseases. New York: D. Appleton 1875. [Author's Preface:] "Fifteen years of uninterrupted activity in the Francis Joseph Hospital for Children in Prague, pursued partly under the guidance of my much esteemed teacher and patron, the Ministerialrath Freiherr Joseph von Löschner, and partly in the independent position of Teacher and Physician in Ordinary to the Hospital, have encouraged me and given me some claim to write this treatise. [...] Prague; August, 1871"
njuffa
Jan 15, 2022 18:36
[From the section on chickenpox in Steiner's compendium; p. 358:]"It is undoubtedly a contagious and transmissible disease, and I have repeatedly communicated it by inoculation, the eruption usually in such cases appearing over the whole surface about the eight day."
 

 Discussion on question by uhoh: Going…

Imported from a comment discussion on scicomp.stackexchange.co...
njuffa
Dec 7, 2021 21:31
@uhoh To avoid misunderstandings: My concern is that the problem size may be too small to make GPU acceleration worthwhile (compared to a vectorized multi-threaded CPU implementation).
njuffa
Dec 7, 2021 21:31
uhoh: Re "The number of atoms could range from 100 to 10,000." I would be concerned about a lack of inherent parallelism. Modern GPUs comprise between 3K (low-end GPU) and 10K (high-end GPU) processing elements working concurrently. Re "In the future I may need to call scipy.special functions" note that special function support in GPU math libraries may not be as extensive as it is on CPUs.
 

 Discussion on answer by T.E.D.: How s…

Imported from a comment discussion on history.stackexchange.co...
njuffa
Nov 5, 2021 13:10
@T.E.D. Is being on open sea defined as being unable to see landmarks on the closest land in the most favorable circumstances? I found some reports that the Turkish coastal ranges are visible from Cyprus on exceptionally clear days. With ancient sailing vessels covering only 5 miles per hour, they probably couldn't complete a sea journey to/from Cyrus while enjoying perfect visibility throughout, though.
njuffa
Nov 5, 2021 13:10
Pascal Arnauld, "Ancient sailing-routes and trade patterns: the impact of human factors", Maritime Archaeology and Ancient Trade in the Mediterranean, pp. 61-80 (online): "Homer himself contrasts the ships and sailing capabilities of the Achaeans with those of the new merchant ships, which ‘crossed the wide gulf of the sea’, while Calypso teaches Ulysses to build a ship in imitation of a merchant ship and to sail it at night, looking at the stars, far from any visible land. Clearly [..] with both day and night sailing on the open sea"
 

 Discussion on question by JonathanRee…

Imported from a comment discussion on history.stackexchange.co...
njuffa
Sep 13, 2021 14:36
massmed.org/About/MMS-Leadership/History/… "[..] in 1855 the legislature enacted laws much more drastic than those of 1792 [..] the heads of all [..] institutions supported in whole or in part by the State were ordered to immediately [...] vaccinate all inmates [..] A parent [..] who neglected to cause his child [..] to be vaccinated before [..] the age of two, unless he had the certificate of a physician [..], was to forfeit $5 for every year during which such neglect continued [..] There was [..] no efficient enforcement [..] in existence."
njuffa
Sep 13, 2021 14:36
Side remark: I recently did some cursory research into mandatory small pox vaccination in the UK (until shortly after WW2) and Germany (until the 1970s). From the sparse data I could find, it appears that the maximum number of people vaccinated as a percentage of total population (over the entire time such laws were in effect) never exceeded 85% for the UK and 80% for Germany. The remainder of the population either claimed various exemptions allowed by law or were simply non-compliant.
 

 Discussion on answer by Raffzahn: How…

Imported from a comment discussion on retrocomputing.stackexch...
njuffa
Apr 17, 2021 17:42
Not everyone may be aware that NMI = non-maskable interrupt.
 

 Discussion on answer by Raffzahn: Pro…

Imported from a comment discussion on retrocomputing.stackexch...
njuffa
Nov 6, 2020 18:47
I was under the (possibly mistaken) impression that the on / off switches I encounter in German-built rental cars when I visit Europe merely trigger a signal to be processed by engine management software running on a micro controller.
njuffa
Nov 6, 2020 18:47
@Raffzahn I think you meant to write "... everything including an on/off switch will need some processor ..." :-)
 

 Discussion on question by Schezuk: Wh…

Imported from a comment discussion on retrocomputing.stackexch...
njuffa
Sep 13, 2020 17:24
@LeoB. I did fairly extensive assembly language programming for x86, SPARC, ARM for about 20 years, until about 2003. Pentium used a simple UV-pipe that allowed for easy optimization by assembly code writers. Even early out-of-order x86 processors had enough implementation artifacts that allowed an assembly programmer to be superior to a compiler. When I examine Clang-generated code today I generally don't see much room for improvement, so I'd say the switchover happened somewhere between gcc 4.0 and early Clang: roughly 15 years ago.
 

 Discussion on question by Granite : I…

Imported from a comment discussion on workplace.stackexchange....
njuffa
Nov 29, 2019 19:17
@Granite Do you have flexible work hours? "get it done in 5 minutes the next morning" suggests you may be a "morning person" who can be very productive early in the day, and therefore should start (and finish) their workday early if possible. I am the opposite: Usually not much of a productive software developer before noon, but I can get quality work done until 9-10pm.
 

 Discussion on answer by Richard Says …

Imported from a comment discussion on workplace.stackexchange....
njuffa
Nov 14, 2019 16:29
@Luaan I agree completely with "You can't write a good question without already being good at debugging". I guess that is one reason so many bad questions get asked (not just on SO).
njuffa
Nov 14, 2019 16:29
[I may be in the same age group as the answerer] There is short-term efficiency and long-term efficiency. Struggling through the work of debugging code (or even a combination of software and hardware!) often leads to valuable insights into the code/system being debugged and a much deeper understanding of it, and generally refines and hones one's debugging skills and intuition. I found that this made me a more proficient and efficient senior software engineer. Lack of debugging skills is a common problem I observe these days.
 

 Discussion on answer by Steve: Can my…

Imported from a comment discussion on workplace.stackexchange....
njuffa
Sep 23, 2019 14:02
USPS offers certified mail with return receipt (a green postcard you receive in the mail, showing date and signature of the recipient of your letter; available for domestic and international service). If the Canadian postal service offers similar service, that is what I would recommend for proof of delivery.
 

 Discussion on question by Janine00: E…

Imported from a comment discussion on workplace.stackexchange....
njuffa
Jan 8, 2019 01:18
@Janine00 Any chance this employee might be harassed or bullied when they sit at their assigned desk, and this is an attempt to stay as far away from the perpetrator(s) as possible?
 

 Discussion on question by Mari-Lou A:…

Imported from a comment discussion on english.stackexchange.co...
njuffa
Dec 19, 2018 17:50
@SvenYargs No wonder: It's as easy as shooting figs in a barrel, and at the same time, more fun than a barrel of money.
njuffa
Dec 19, 2018 17:50
PETA discovered there is 'more than one way to skin a cat'. Years ago, I read that someone suggested to use 'more than one way to skin a kumquat' instead.
 

 Discussion on question by Jared: My b…

Imported from a comment discussion on workplace.stackexchange....
njuffa
Aug 23, 2018 17:07
@Jared Could you add a tag to clarify what country or region this is? Cultural expectations may differ.
 

 Discussion on question by Dexter M: H…

Imported from a comment discussion on workplace.stackexchange....
njuffa
Mar 26, 2018 15:46
@Dexter M Is this startup experience generally valuable to you (gaining a lot of experience, good pay, + stock options)? Everybody has different limits. It sounds like you are generally on good terms with your boss. Have you tried a one-on-one with your boss to make it clear what your limits are? This has worked for me in the past: It was at the last startup I did in my career, when I was about 40; I could sustain about 60 hours of productive work per week (including work on Saturdays) for more than a year using a very disciplined lifestyle.
 

 Discussion on question by detcader: H…

Imported from a comment discussion on workplace.stackexchange....
njuffa
Feb 8, 2017 22:29
@detcader But cars are a commodity, and therefore I would avoid quoting salary surveys (and never have used them; my preferred tool to find out my "market value" was to interview relatively frequently). What counts is how valuable you are to this organization. The value someone else brings to this organization may well be very different, as well as the value you bring to some other organization.
 

 Discussion on question by Overworker:…

Imported from a comment discussion on workplace.stackexchange....
njuffa
Mar 14, 2016 18:01
@Overworker Is it possible for you to be promoted (in the technical track)? I worked with similar intensity (7 days a week, 70 hours / week) early in my career for similar sized companies, and either of two things happened within a two-year time span: (1) The company gave me a promotion to a higher pay grade, e.g. MTS to Senior MTS (2) I switched companies, getting on average a 20% higher salary in the process. For promotion it is important that you don't just do lots of work, it must be the right kind of work (e.g. create intellectual property, informally lead junior team members).