PodSearch

Under the Radar

185: Advanced Math

 

00:00:00   Welcome to Under the Radar,

00:00:01   a show about independent iOS app development.

00:00:04   I'm Marco Arment.

00:00:05   - And I'm David Smith.

00:00:06   Under the Radar is never longer than 30 minutes,

00:00:08   so let's get started.

00:00:09   - So Dave, I don't know about you,

00:00:11   but in school, I was really good at math,

00:00:14   only up to a certain point.

00:00:18   Maybe like middle of high school.

00:00:20   Once we started getting into more advanced math topics,

00:00:24   like I think for me the boundary was calculus.

00:00:27   I didn't take calculus till college,

00:00:28   and I was terrible at it, and I failed it twice,

00:00:32   and never went and retook it,

00:00:33   and never really learned calculus.

00:00:35   I failed linear algebra also twice,

00:00:37   doing advanced matrix operations and stuff like that,

00:00:40   and never retook that, never fixed that.

00:00:44   Anything involving really advanced calculations like that,

00:00:47   that wasn't just basically combinations of basic,

00:00:51   the basic four calculator operations,

00:00:53   or stuff like that.

00:00:55   I was very bad at, or never learned,

00:00:58   or tried to learn and failed at.

00:01:00   And yet we work in a field where our tools

00:01:04   are super fast math machines that do what we tell them to,

00:01:08   and that we often need math for.

00:01:12   And I think it's funny that a lot of programming

00:01:15   does indeed need math, but it's mostly the simpler math

00:01:19   that I was really good at already.

00:01:20   It's mostly elementary, middle, and maybe high school math.

00:01:24   But it's not, most programmers don't have to do things

00:01:27   like calculus or linear algebra, or theory proofs

00:01:31   and everything like that in their regular work,

00:01:33   or any of their work really.

00:01:34   I think those are specialties.

00:01:36   But I do occasionally run into some need

00:01:39   for a better understanding of math than what I have.

00:01:45   And I'm curious, does this happen to you too,

00:01:48   and how do you solve it?

00:01:50   - Yes, so A, it happens all the time.

00:01:53   And it is like one of the,

00:01:56   it's almost like a running joke in my house,

00:01:57   where all of a sudden, it's sort of like a scene

00:01:59   out of Beautiful Mind, and my desk just becomes

00:02:01   just strewn with papers, filled with scrawled formulas

00:02:05   all over the place, because I'm not particularly,

00:02:08   yeah, I had a similar experience with math.

00:02:10   I can do, through Algebra II trig,

00:02:13   was sort of like, I think I took,

00:02:17   that was the last level of math where,

00:02:20   and I feel like that's the level of where math

00:02:22   is still intuitive to some degree,

00:02:25   but it's based on tangible, understandable things,

00:02:29   in a way that like addition, when you teach it to a child,

00:02:32   is like, I have one apple, and I get another apple,

00:02:36   how many apples do I have?

00:02:37   It's like, well, you have two.

00:02:39   And there's an intuitive and like a,

00:02:42   something concrete that you can wrap your head around.

00:02:44   And then you get to a certain point in math,

00:02:46   and it stops being that.

00:02:49   And it becomes this much more abstract,

00:02:53   sort of theoretical thing that I can sort of understand,

00:02:57   that in calculus, what you're doing is you're calculating

00:03:00   the area under a curve, or the slope of a line.

00:03:04   That is fundamentally what most of calculus is doing.

00:03:08   But just knowing that that's functionally what you're doing,

00:03:12   actually how you go about doing that,

00:03:15   very quickly becomes this totally abstract concept,

00:03:19   where how do you know what the slope of a sine curve is?

00:03:22   And it's like, well, that's like through deep magic.

00:03:26   I don't know.

00:03:26   All I learned at a certain point was that

00:03:29   this is the formula, and I can apply this formula,

00:03:32   and I can get the right answer.

00:03:33   But I never had a intuitive understanding

00:03:36   of anything beyond, yeah, that's sort of like

00:03:38   that algebra two trig area.

00:03:40   Like trigonometry is where it starts to kind of,

00:03:43   I'm hitting the limit where I understand that I'm like,

00:03:46   the ratio between two lengths of a triangle,

00:03:51   that that ratio is related to the angle of the corner,

00:03:54   and things like that.

00:03:55   Like it's sort of concrete and sort of understandable,

00:03:58   but beyond that, it kind of falls apart.

00:04:00   But this happens all the time.

00:04:03   And I feel like often what is most kind of frustrating

00:04:07   is that this math shows up in places where you can,

00:04:12   it's like at this boundary between a feature being

00:04:15   like all right, and a feature being really cool.

00:04:18   Like there's this, if you wanted the extra flourish,

00:04:20   if you wanted to be that little bit cooler,

00:04:22   or do a little bit, like something a little bit fancier,

00:04:25   is where the math starts to show up.

00:04:27   Where the straightforward way is, you see,

00:04:31   it's like that's when you're dealing with just like

00:04:32   simple rectangles and moving them around

00:04:35   in straightforward way.

00:04:36   But if you want to have like a cool curve

00:04:38   when you're doing an animation,

00:04:39   or you want to make the shape of something

00:04:41   not just be plain, but be sort of advanced,

00:04:46   or even moreover what I've been running into

00:04:49   a lot recently in my work is,

00:04:51   there's a naive solution to a problem,

00:04:54   and then there's the accurate solution to the problem.

00:04:58   And the naive solution is really straightforward

00:05:00   and easy to work with.

00:05:02   But the actual solution that is like sort of correct

00:05:05   in physics or correct in detail is where all of a sudden

00:05:09   now I need to pull out all of the math

00:05:11   and go and spend a lot of time just like researching online.

00:05:15   And a lot of it is the unfortunate kind of research

00:05:17   where I feel like I solve this problem a lot

00:05:20   by trying to find, learn just enough

00:05:23   that I can find the right formula

00:05:25   that I can apply and get the right answer,

00:05:27   even if I don't understand what I'm doing.

00:05:30   Like there's a ton of the math that I was using

00:05:32   for like my moon project where it's like,

00:05:35   I don't understand, I have no idea why

00:05:38   like this equation gives me the, you know,

00:05:40   the longitude of the moon right now.

00:05:43   But it does.

00:05:44   And it works.

00:05:45   And once I can like, I'm just like empirically verifying it

00:05:48   and then going like finding, you know, the NASA table

00:05:51   where it's like, you know, I can get a table

00:05:53   that forecasts the values of certain things

00:05:55   over different times.

00:05:56   My values match their values like plus or minus

00:05:59   a small error percent, and then I'm good.

00:06:01   And I don't need to understand why the formula works

00:06:04   or what's going on necessarily, but like that is,

00:06:06   that is a situation that has happened

00:06:08   so often to me recently.

00:06:10   - Yeah, it happens to me probably every few months

00:06:13   where I will be tackling a feature

00:06:15   or trying to figure out, you know, some weird edge case

00:06:18   or something where I will quickly realize like,

00:06:21   you know what, I could use more advanced mathematical

00:06:24   knowledge than what I have to solve this.

00:06:25   Or I need more advanced knowledge to solve this.

00:06:28   And for me, like you know, like another big hole

00:06:29   in my knowledge is most kind of advanced stats

00:06:33   and like data analysis.

00:06:35   Like I can't do like data mining.

00:06:37   I still don't understand machine learning at all.

00:06:40   And I'm almost, I haven't really tried it,

00:06:43   but you know, there's huge areas of that

00:06:45   where I feel like I either have fallen behind

00:06:48   or was never even in the race that I probably

00:06:51   should familiarize myself with.

00:06:53   But even, you know, basic stuff like if I'm trying

00:06:55   to figure out a pattern in a bunch of data I have,

00:06:58   and I'm trying to figure out like yeah,

00:06:59   I can calculate like the average and stuff.

00:07:01   But if I try to go beyond that, to be like all right,

00:07:04   what if I like, you know, try to find clusters

00:07:06   of meaningful data or I try to find, try to, you know,

00:07:09   rule out the outliers and you know, focus only

00:07:12   on the middle, like that's all like much more

00:07:15   advanced math than anything I ever actually

00:07:17   officially learned.

00:07:18   And sometimes you can take advantage of libraries

00:07:21   or functions or formulas you can find online

00:07:23   to do it for you.

00:07:25   Sometimes you just kind of fudge it and you kind of like,

00:07:27   all right, well, I guess I'll figure out how to do this

00:07:30   myself with the knowledge I have without actually

00:07:33   looking up the right way to do it.

00:07:35   And sometimes you can get an approximate solution

00:07:36   that kind of works.

00:07:37   Sometimes you accidentally come upon the actual solution

00:07:40   without even realizing it, although that's less common.

00:07:43   And I find that like a lot of times the kind of like

00:07:48   half fake hack solution that you kind of stumble upon

00:07:52   by doing it naively is often good enough

00:07:55   and is often fast enough.

00:07:56   And in fact, there's lots of methods that you can find

00:08:00   online of like, of you know, major apps or especially games.

00:08:03   Games often take a lot of shortcuts for performance

00:08:06   where they did some kind of rough approximation

00:08:09   of something and that proved to be good enough

00:08:13   or close enough for their purposes.

00:08:15   And that saved them either from having to know

00:08:19   a more advanced version or saved the computing time

00:08:21   of the computer having to do a more advanced version.

00:08:24   - Yeah, and that good enough is such a, like,

00:08:26   I feel like in weird ways, like, 'cause part of me

00:08:30   as a developer, I enjoy solving like the hard problem

00:08:33   and doing it well.

00:08:34   Like, that's part of what I like doing.

00:08:37   But it's so often the, like, is this, it's like the increase

00:08:41   in effort to have the correct answer versus a close answer

00:08:45   is, you know, is wildly disproportionate.

00:08:50   Like I was thinking recently, I was doing a thing.

00:08:52   So I, you know, I had, if you imagine you're trying to show

00:08:55   the color and associate it with a temperature, you know,

00:08:58   so this is for like a weather display.

00:09:00   Like in some ways, like the first version of this that I did

00:09:04   was a, like I just did the super naive version

00:09:07   where if you take the hue of like, you know,

00:09:10   it starts like a hue zero or in, you know, in UI color.

00:09:15   And that's red.

00:09:18   And if you go up to about like 220-ish, that's like blue.

00:09:23   And in between it goes like red, orange, yellow, green

00:09:27   to blue, like it's, you know, sort of as you would expect.

00:09:29   And the super simple version I did initially was just like,

00:09:32   well, I'll take like the temperature in,

00:09:36   and say I do it like in Fahrenheit from zero to 120,

00:09:40   work out like what percent of that is,

00:09:42   and then just like map that onto the hue.

00:09:44   And so when it's really hot, it'll be, you know, red.

00:09:47   And then when it's really cold, it'll be really blue.

00:09:50   And like that sort of works.

00:09:51   But then I started, you start to get like,

00:09:53   well, what if I wanted to make it a little bit like,

00:09:55   like have a more consistent visual appearance to it

00:09:59   or have a more, like a thoughtful,

00:10:02   or like I could have different points along the way.

00:10:05   Well, then I need to start to be getting

00:10:06   into like color blending.

00:10:07   And color blending is this whole crazy,

00:10:10   like you see one of these things,

00:10:11   like you open the first box,

00:10:13   like you pick the book off the shelf that says color blending

00:10:16   and you're like, oh, that's interesting.

00:10:17   And you can just like, the first version is just do like,

00:10:19   you know, I pick the points along the curve

00:10:21   that I want the colors to be,

00:10:23   and then I'll just do like a linear interpolation

00:10:24   between like the RGB values at each point,

00:10:27   which like mostly works.

00:10:30   But then you can have these weird like middle,

00:10:31   mid points in between your interpolations

00:10:33   where like the color isn't really what you want,

00:10:35   but that is like the color that you just asked for,

00:10:38   where you were, what you really start to get into,

00:10:40   like then you get into like color perception

00:10:42   and how people are sensitive to different colors

00:10:44   at different intensities.

00:10:47   So, so like a yellow and a blue and a red,

00:10:51   like at a constant hue, if you keep the saturation

00:10:54   and like the brightness the same,

00:10:56   like one of them will look darker than the other.

00:10:58   And so you can end up with these weird cases

00:11:00   where if you're just blending between them,

00:11:02   like it looks as though the color is getting like brighter

00:11:04   and darker in weird and uncomfortable ways.

00:11:08   And that's because of the way people perceive color.

00:11:11   That is, you know, way,

00:11:13   all of a sudden you're starting to get like,

00:11:14   even beyond math, you're starting to get into like biology

00:11:16   and the way that the rods and cones in your eyes work.

00:11:19   And it's like, at a certain point, it's like, okay,

00:11:21   I'm just gonna close this book

00:11:23   and I'm gonna start with the simple version again,

00:11:26   because the effort to get it perceptually correct

00:11:29   has probably now over, like exceeded the actual increase

00:11:33   in quality that I would get by opening up this math.

00:11:36   And like, that's an example where like,

00:11:37   I ended up going with a fairly simple version

00:11:40   that kind of mostly works

00:11:41   and doesn't deal with all the weird image perception issues,

00:11:44   but gives me a pretty good answer

00:11:46   that looks pretty good most of the time.

00:11:49   And it's like, okay.

00:11:50   Or you can do the other classic thing where in programming,

00:11:53   where I find like I often will take,

00:11:55   like there's a weird place that my 80% solution works,

00:12:00   but except for like in this, you know,

00:12:01   in this temperature case,

00:12:02   it's like except for between 60 and 70 degrees,

00:12:06   my color looks wrong.

00:12:07   And so I just like have an if statement

00:12:09   that if the temperature is between 60 and 70,

00:12:12   change it in this way.

00:12:13   That's not like mathematically correct.

00:12:15   It's not like this beautiful, smooth curve.

00:12:17   And if you really look closely between like, you know,

00:12:19   this like one degree, one way, one degree the other,

00:12:22   you can notice this, you know, discontinuity in the curve.

00:12:25   It's like, you know, like you could,

00:12:27   but I only ever display like one color

00:12:29   on the screen at a time, so you know, it's probably fine.

00:12:33   And like that's this kind of,

00:12:34   this place where you find where you're trying to like,

00:12:36   where on this math spectrum do I actually wanna be?

00:12:39   Like do I wanna go like go crazy down the rabbit hole

00:12:43   and then come up on the other side

00:12:44   with this perfect solution that is like technically correct?

00:12:48   Or do I want something that's, you know,

00:12:49   mostly good and looks fine?

00:12:52   - I mean, and you know, an example like that,

00:12:54   like you could even do something even dumber and simpler,

00:12:57   you just have a lookup table

00:12:58   for like integer temperature values.

00:13:00   Because does it really matter

00:13:02   if you have a slightly different color

00:13:05   for 60 degrees versus 60.5 degrees?

00:13:08   - Sure.

00:13:09   - Like it probably doesn't matter that much.

00:13:10   I bet nobody would ever notice.

00:13:12   And so you could actually just like run the basic formula,

00:13:15   get the colors, map them to a lookup table

00:13:17   for each temperature scale, and then just look at it

00:13:19   and say like, which of these look wrong?

00:13:20   And then just manually adjust the ones

00:13:22   like for those buckets.

00:13:24   And it would be a heck of a lot faster in performance.

00:13:26   It wouldn't really cost you much memory.

00:13:28   And it would actually be a reasonable solution

00:13:31   to that problem since in this particular application,

00:13:34   you don't need a smooth blending between intervals.

00:13:38   Because no one is dragging a slider around,

00:13:41   adjusting the temperature in a smooth floating point domain

00:13:45   and making it so that they have to see

00:13:47   these nice little gradients going.

00:13:49   And like you actually could do

00:13:50   something even simpler than that.

00:13:52   I feel like there's a lot of cases like that

00:13:54   where we have opportunities for simplification

00:13:58   or optimization like that that we don't think about

00:14:01   'cause we want like the theoretically best solution.

00:14:03   Like this function, it won't work for floating point values

00:14:08   in this certain range here.

00:14:09   But it's like okay, well do you ever have those?

00:14:11   Or does it matter if you just round those

00:14:13   to the nearest integer first and then do the operation?

00:14:15   And in a whole bunch of applications,

00:14:17   the answer is no, it doesn't matter.

00:14:18   Just make it work for what you will ever actually see.

00:14:23   And if it doesn't quite look right

00:14:24   or doesn't quite draw right for these really weird edge cases

00:14:28   that literally your app will never, it should never happen,

00:14:31   it doesn't really matter

00:14:32   whether it was theoretically right or not.

00:14:34   - Yeah, and the lookup table example I think is a great one.

00:14:36   Like I also, it makes me think of,

00:14:38   with all my moon calculations,

00:14:40   at one point what I was trying to do

00:14:42   is show the timing of the new and full moons in the app.

00:14:48   So you can say like essentially jump,

00:14:50   but you can kind of jump forward in time

00:14:52   and know like when is the actual next full moon gonna occur?

00:14:56   And I went down the rabbit hole of trying to work out

00:14:58   like how do you calculate that?

00:15:00   And it turns out it is just like, it's not even a book.

00:15:03   It's like a collection of books and 17 people's lives

00:15:07   to try and understand how you calculate when this is.

00:15:11   And I go down this road for like maybe half a day.

00:15:13   And in the end, I just like realize,

00:15:16   okay, NASA has published a list of the full and new moon

00:15:21   for the next hundred years.

00:15:24   I'll grab like 10 years of those

00:15:25   and just take the number that NASA has calculated

00:15:28   and just hard code it in the app.

00:15:30   And if I wanna know when the next one is,

00:15:31   I'll just see like when's the next,

00:15:32   when's the one greater than my current one?

00:15:35   And I just have, I think I go 10 years into the future

00:15:38   and it's like, well, in 10 years,

00:15:39   the app will stop working in that particular way.

00:15:41   And either I will have fixed it and extended it

00:15:44   or the app won't exist or I won't care.

00:15:46   And like that, like, or similarly,

00:15:50   I did the same thing for like lunar eclipses,

00:15:52   which is like something like,

00:15:53   it worked out well in that particular case

00:15:55   'cause like lunar eclipses, there aren't one for a while,

00:15:57   so I don't need to worry about that particular feature.

00:15:59   But when it does become our niche,

00:16:00   I'll just hard code the values

00:16:01   because I know when they're supposed to happen.

00:16:03   Rather than like trying to calculate

00:16:06   based on your current coordinate,

00:16:08   what the, when you would perceive one,

00:16:09   which is like a theoretically possible math problem.

00:16:13   But like just hard coding the value, that worked just as well.

00:16:16   - We are sponsored this week by Linode.

00:16:18   Whether you're working on a personal project

00:16:19   or managing your enterprise's entire infrastructure,

00:16:21   Linode has the pricing, support, and scale you need

00:16:24   to take your project to the next level.

00:16:26   They have 11 data centers worldwide,

00:16:28   including their newest data center in Sydney, Australia.

00:16:31   With their enterprise-grade hardware,

00:16:32   S3-compatible storage option,

00:16:34   and their next generation network,

00:16:35   Linode delivers the performance you expect

00:16:37   at a surprisingly good price.

00:16:39   I use Linode, Dave uses Linode,

00:16:41   I gotta say I am super happy with it.

00:16:43   I've been with them for almost a decade.

00:16:44   I have all of Overcast hosted there,

00:16:46   I have my personal site server hosted there.

00:16:48   It's just wonderful.

00:16:50   You can get started today on Linode with a $20 credit.

00:16:53   You get access to native SSD storage,

00:16:55   their 40 gigabit network, industry leading processors.

00:16:59   You get root access to your server

00:17:00   along with an API version four now

00:17:02   and a Python CLI tool to automate setup if you want to.

00:17:05   Their nano plans start as low as $5 a month,

00:17:08   and they have lots of stuff above and beyond that

00:17:10   if you have higher needs.

00:17:11   Including things like dedicated CPU plans

00:17:13   with physical cores reserved just for you.

00:17:15   GPU compute plans, a whole bunch of other stuff.

00:17:17   So, go to linode.com/radar

00:17:21   and use promo code radar2020

00:17:23   when creating a new Linode account.

00:17:24   You get a $20 credit towards your next project.

00:17:26   And by the way, they're hiring.

00:17:27   If that interests you, go to linode.com/careers.

00:17:30   Anyway, that URL for the $20 credit one more time

00:17:33   is linode.com/radar, promo code radar2020.

00:17:37   Our thanks to Linode for their support of this show

00:17:40   and all of Relay FM.

00:17:41   - So I've got to imagine with a lot of your sound processing,

00:17:45   like there's gotta be a huge amount of math

00:17:48   and like even, I mean at a certain point,

00:17:50   like super complicated math,

00:17:52   where you start to get into like Fourier analysis

00:17:54   and the like things that are very beyond what I,

00:17:58   I don't really even know which they involve,

00:18:00   but I've gotta imagine things like Voice Boost 2

00:18:02   that you just were working on.

00:18:04   Like that's gotta involve a lot of this kind of like,

00:18:07   there's a lot of theoretical math

00:18:08   that you have to deal with.

00:18:09   Is that something that you have to like deal with

00:18:11   the actual like formulas and equations for

00:18:13   or is it, at a certain point it becomes very like empirical

00:18:16   and you're just comparing numbers

00:18:18   and like multiplying and dividing them as needed?

00:18:21   - It's interesting, like I,

00:18:23   and I have no formal training in signal processing

00:18:27   or anything like this and so,

00:18:29   and what I have to deal with when processing sound

00:18:31   is a lot of signal processing stuff.

00:18:33   You know, it's stuff like filters and you know, bi-quads

00:18:38   and yeah, you mentioned that the FFT, you know,

00:18:40   stuff like that.

00:18:41   I'm actually lucky that Overcast currently does not use

00:18:43   an FFT for anything.

00:18:44   The only time I've ever actually used one in Overcast

00:18:46   was in the very early versions when I had like

00:18:49   the peak meters that were frequency sensitive,

00:18:51   the animatability artwork.

00:18:53   - Sure.

00:18:54   - I don't do that anymore 'cause it was a little bit

00:18:56   expensive for performance and it looked a little bit

00:18:58   intrusive on the screen.

00:18:59   People didn't like it as much as I thought they would.

00:19:01   So now I just do the little peak meters and the pause button.

00:19:04   Yes, I know in the current app store version they're gone.

00:19:06   Don't worry, they're coming back.

00:19:07   They're in the beta right now.

00:19:09   And that's just doing amplitude so I don't have to separate

00:19:11   the frequency so it's much simpler.

00:19:13   But I've had to learn a lot of DSP stuff.

00:19:17   And fortunately, most of it is like you learn a couple

00:19:23   of concepts like how you can't just apply different gain

00:19:28   per sample as it's curving and expect it to sound correct

00:19:31   'cause it won't.

00:19:32   Or you know, things like doing basic look ahead,

00:19:35   doing like attack release timing.

00:19:38   But most of that is a very, very large amount of very,

00:19:41   very simple map operations.

00:19:43   It's you know, multiplication, it's addition.

00:19:46   One of the only weird things was that sound is perceived,

00:19:51   as you mentioned, like with the color example,

00:19:53   like the way it's represented is with you know,

00:19:56   regular values from one to zero to negative one

00:20:00   as it goes like up and then back down like when you see

00:20:02   the waveform graph.

00:20:04   So it's like one to negative one in the float domain

00:20:06   and with zero in the middle.

00:20:07   But the way it's actually perceived by humans is not linear,

00:20:11   it's logarithmic because of the decibel scale.

00:20:15   And so I do a lot of operations in the decibel domain

00:20:19   to make them sound better.

00:20:20   And this is one of those examples where like,

00:20:23   it doesn't, I don't really need to do it in the decimal,

00:20:26   in the log domain.

00:20:27   I could do it in the linear domain and it would sound okay

00:20:30   and most people wouldn't notice.

00:20:31   But because we perceive sound in the log domain,

00:20:35   the more I could do in that domain, the better it would be.

00:20:38   But also, logarithms are a complicated operation

00:20:41   for computers to do.

00:20:42   And so you gotta be careful like not to do too many

00:20:44   conversions back and forth or not to try to like

00:20:47   interpolate the log at every single point of something.

00:20:49   Like there's all sorts of optimization potential.

00:20:51   And there's also things like approximation functions.

00:20:55   Where like, there are fast logarithm approximation

00:20:59   algorithms out there that like, they get you kind of

00:21:02   part of the way there.

00:21:03   And sometimes that's good enough.

00:21:05   Sometimes you know, if you have like,

00:21:07   if your approximation is good to like three decimal places,

00:21:10   like that's actually pretty good for your purposes.

00:21:12   In some context it's not.

00:21:14   And so there was a whole bunch of figuring that kind

00:21:19   of stuff out.

00:21:20   But for the most part, most of the math I'm doing is like,

00:21:23   vector based, which is like, you know,

00:21:26   basically look at the accelerate library

00:21:27   and the VDSB functions that Apple provides.

00:21:30   And it's mostly just applying the basic operations,

00:21:35   like addition, subtraction, multiplication,

00:21:37   to vectors of floats, which is just a fancy word

00:21:40   for arrays of floats.

00:21:41   So you have like a float array of like a thousand floats

00:21:44   and alright, these are the next thousand samples.

00:21:45   And you can say alright, what's the maximum magnitude

00:21:48   of, in this thing, which is like, you know,

00:21:51   basically the absolute value maximum.

00:21:53   And so you can say alright, that's the peak.

00:21:55   So when I'm rendering my peak samples,

00:21:57   I can say alright, take a look at this next block

00:22:00   and what's the maximum magnitude, VDSB max m g v, bam.

00:22:04   These function names make no sense when you're first

00:22:07   looking at them, but as you get used to them,

00:22:08   you kind of pick up the convention.

00:22:10   There is a convention.

00:22:11   And so then you can rattle them off and sound really crazy.

00:22:16   And then so you can call these functions on them

00:22:18   and it basically like rips right through it.

00:22:20   It's like these functions are so fast.

00:22:22   And then I have the maximum amplitude value,

00:22:24   and maximum absolute amplitude value in the next block

00:22:27   of samples and that gives me my peak value

00:22:29   for my peak meter display.

00:22:30   So like the concepts and learning what these do

00:22:35   looks intimidating from the outside because you have

00:22:36   all these like mysteriously named functions

00:22:39   and you're operating on very large amounts of data,

00:22:42   but really what you're doing is very simple operations.

00:22:44   And the challenge for me was more in learning

00:22:48   like how to manipulate sound to do what I wanted

00:22:51   in like basic techniques, but it wasn't so much about

00:22:54   like the math behind it.

00:22:56   The only exception to that is I had to do some amount

00:23:00   of figuring out bi-quads, which is like a method of,

00:23:03   if you ever want to implement like an EQ or frequency

00:23:07   modification, the way I thought naively that you did it,

00:23:11   I assumed that you would use an FFT to transfer

00:23:15   from the time domain to the frequency domain,

00:23:18   modify like whatever bucket you wanted a frequency,

00:23:21   is like change this value from you know,

00:23:23   .7 to .6, that'll drop that frequency,

00:23:25   and then do an inverse FFT and put it back.

00:23:27   And it turns out that isn't how you do it at all.

00:23:30   You can do things that way, but it's much more complicated

00:23:33   and way slower and you have to do a lot of other things

00:23:36   to make it not you know, suck or distort the sound

00:23:39   in weird ways.

00:23:40   And so you can also use a thing called a bi-quad filter

00:23:43   that is way simpler and this is what everything does.

00:23:45   Like pretty much any kind of like EQ or filter thing

00:23:49   that you see in apps, the iTunes EQ,

00:23:52   all that stuff is just bi-quad filters.

00:23:54   And I had no idea what those were or how they worked.

00:23:57   And I still don't.

00:23:59   I like, all I know is that I found some equations

00:24:02   for how to generate like the numbers to plug into

00:24:05   the bi-quad filter and there's a VDSB bi-quad function

00:24:09   that does the calculation for me.

00:24:11   So all I did was like how do I make these coefficients?

00:24:14   And I looked up you know, certain formulas,

00:24:16   alright here's how you make a notch filter,

00:24:18   here's how you make a peak filter, like stuff like that.

00:24:21   And then I plug it into this formula that's already there.

00:24:24   So that's an example where like there is some very complex

00:24:27   math behind that that I probably should know

00:24:31   but I didn't have to 'cause the library took care

00:24:33   of most of it for me and some you know, a few days of

00:24:37   you know, web research and Stack Overflow and Google

00:24:39   and stuff took care of the rest.

00:24:42   - The thing you said though like the you probably should know

00:24:45   like I think increasingly I don't,

00:24:48   I'm not sure that's even true.

00:24:49   - Hmm, that's a good point.

00:24:50   Like 'cause I think there's the, this is the like,

00:24:53   this is the root of this problem is it's like,

00:24:55   do you need to like, do you need to understand

00:24:59   the physics of leverage in order to use a hammer?

00:25:01   You know like, would you be more, would you be more capable

00:25:07   or wield it more efficiently if you understood the way

00:25:11   of like, of the way that a lever works and like all of

00:25:14   and like that type of or like the transfer of inertia?

00:25:16   Like there's these things where if you understand

00:25:19   the fundamental, like the underpinnings and the math

00:25:23   and the physics and whatever it is of what's going on,

00:25:25   like you can imagine and I can sort of, I can imagine

00:25:30   myself in this pretend, like this pretend world

00:25:32   where I understand this thing that I, my,

00:25:35   the end result would be better.

00:25:36   And like it probably would be better but almost certainly

00:25:40   not by as much as you would imagine it is.

00:25:43   That like I think increasingly what I find is it's like

00:25:45   if I can understand something enough to know how to use it,

00:25:49   and understand that there might become a point where

00:25:52   there's a bug or an issue or an edge case that I can't

00:25:54   deal with, like that's fine.

00:25:57   But it also means that I don't have to spend like three

00:25:59   years getting a PhD in the subject to like,

00:26:02   because I think what I realize is you hit a point where

00:26:06   the like, there's this massive chasm between the just like

00:26:10   basic intuitive understanding of something and then like

00:26:13   the actual deep understanding.

00:26:15   And it isn't just like, oh, this is like, you know,

00:26:18   a quiet afternoon sipping a cup of coffee on my porch

00:26:21   reading a book.

00:26:22   It's like, no, this is like, I need to go to like college

00:26:25   level classes to like unpack and understand what's going on.

00:26:29   You know, it's like, I don't really understand how

00:26:31   Bezier paths work, like the math behind them.

00:26:35   Like I remember trying to kind of get a sense of what's

00:26:37   going on there.

00:26:39   And it's like, I sort of could maybe get like, I was like,

00:26:42   you know, could crack through like 0.1% of it.

00:26:45   But at a certain point, I'm like, I don't know.

00:26:47   And it's fine.

00:26:48   I just like, you know, move the control point until it looks

00:26:51   correct, and then that's fine.

00:26:54   And like, if I truly understood the math behind it,

00:26:57   maybe that would be better.

00:26:58   But I get a certain point.

00:27:01   Like, I think the key thing in some ways that in with,

00:27:04   like in my relationship with math and my work is

00:27:06   understanding that like, it's, I was there were problems

00:27:09   that I was not solving, or like solutions I was ignoring

00:27:13   because I thought I would have to understand it at a deep

00:27:16   level in order to use it.

00:27:18   And I think increasingly, it's like you, you don't need to

00:27:21   necessarily understand it, you can, it's totally fine and

00:27:24   legitimate to just have a working version of the solution.

00:27:28   And then use that, you know, there's many things in computer

00:27:32   science and in, you know, in the design of applications that

00:27:35   I don't understand.

00:27:36   And I don't need to understand like, I'm not really good on

00:27:39   some of the graphics stuff where like, how, like,

00:27:41   wave frame, frame buffers work, and the display links.

00:27:45   And like, there's these things, these tools that I use when

00:27:47   I'm like making an animation that I really don't understand

00:27:51   that, you know, that the ins and outs of, but I understand

00:27:54   enough to make it work.

00:27:56   And I think that like accepting and being okay with that

00:27:59   is, opens up all these problems to you that you just can,

00:28:03   you can solve now.

00:28:03   Like, cause if you, if you had to understand, what was it,

00:28:06   bicubic?

00:28:07   - Biquads.

00:28:08   - Biquads.

00:28:09   - Yep.

00:28:10   - If you had to understand biquads in order to use biquads,

00:28:12   then you probably wouldn't be using biquads and you wouldn't

00:28:15   have an EQ filter and you wouldn't have features and things

00:28:18   that you know, that would be available to you.

00:28:20   And so I think increasingly I find like, just, it's okay.

00:28:24   Just accept that this is someone else's job in the same way

00:28:27   that I don't design Silicon chips, I just use them.

00:28:30   Like someone else has done the hard work of coming up with

00:28:33   the math and you know, they sure they got a PhD out of it.

00:28:36   And they can put some letters after their name as a result,

00:28:38   but I just want to, you know, use the thing and have it work.

00:28:42   And that's probably good enough for me.

00:28:45   - I guess I usually do feel that same way.

00:28:47   That is usually my philosophy of like, you know,

00:28:49   somebody else can figure out the complex mathematical

00:28:53   details of this thing and I can just use it.

00:28:55   It does occasionally bite you in the butt, like if something

00:28:58   goes wrong or if you're having some kind of massive

00:29:00   performance issue that like, and it does help to understand

00:29:04   some of the underlying stuff then, but that just doesn't

00:29:07   happen very often anymore.

00:29:08   So you actually can just like, let someone else do the

00:29:11   algorithm or the formula, figure it out, and then just

00:29:14   use it.

00:29:15   - Yeah, I think then that's good enough.

00:29:17   And like, it's fun to use math that I learned in high school

00:29:21   that at the time when I was learning it, I thought,

00:29:23   why would I ever use this?

00:29:24   You know, I'm never gonna use trigonometry in my real life.

00:29:27   And it's like, actually I use trigonometry all the time.

00:29:29   - Yeah, me too.

00:29:29   - So if there's any kids listening, it's like trigonometry

00:29:32   is the real stuff, make sure you learn that.

00:29:34   - Yeah, right triangle trig, that's important.

00:29:37   - That's important.

00:29:37   - Not much after that is, but that's important.

00:29:39   - That's important.

00:29:40   Forget about calculus, right triangles is where it's at.

00:29:43   Thanks for listening everybody, and we'll talk to you

00:29:45   in two weeks.

00:29:46   - Bye.

00:29:47   [