Logs: liberachat/#haskell
| 2021-07-29 18:11:11 | <motle> | roboguy_: well, they dont use tails, its all index edge stuff adjacency matrix no good imo |
| 2021-07-29 18:11:26 | × | Null_A quits (~null_a@2601:645:8700:2290:44f7:81a6:341:7abe) (Remote host closed the connection) |
| 2021-07-29 18:11:27 | <motle> | wants proper hetrogenous like |
| 2021-07-29 18:12:06 | <roboguy_> | I mean the stuff you want to replace with stencil convolution |
| 2021-07-29 18:12:10 | <motle> | you know what i mean, making it all isomorphic to grpahs with hetrogenous list branches with explicated cycles |
| 2021-07-29 18:12:27 | <motle> | roboguy: yeah, things like repa |
| 2021-07-29 18:12:37 | <motle> | parallel zips iirc |
| 2021-07-29 18:12:53 | <motle> | and the indexing on the fast random access containers |
| 2021-07-29 18:12:56 | <motle> | something about idiom |
| 2021-07-29 18:13:10 | <motle> | i want an actual extensible datatype |
| 2021-07-29 18:13:21 | <motle> | the HList cyclic tree |
| 2021-07-29 18:13:33 | <motle> | everythings just that anyway |
| 2021-07-29 18:13:33 | <lechner> | Hi, will 'show' convert a UTF-8 ByteString into a String with Unicode code points? |
| 2021-07-29 18:13:36 | <roboguy_> | hmm, ok then I think I really don't understand, hah. I haven't seen anyone use repa to implement program evaluation |
| 2021-07-29 18:14:00 | <[exa]> | lechner: utf will usually get shown as \u1234 or something |
| 2021-07-29 18:14:01 | <motle> | lol, stencil convolutions |
| 2021-07-29 18:14:10 | <motle> | matrix stuff, linear algebra |
| 2021-07-29 18:14:17 | <motle> | tensors like applicative |
| 2021-07-29 18:14:59 | <motle> | you would have a faster implementation if it was block diagonal with 0 elsewhere, typical convolution |
| 2021-07-29 18:15:01 | <[exa]> | motle: someone talked about implementing similar stuff here a while ago, actually I think he foudn some solution, you might search the logs |
| 2021-07-29 18:15:19 | <roboguy_> | motle: Have you seen Conal Elliott's papers on automatic differentiation and generalized convolution? |
| 2021-07-29 18:15:32 | × | Atum_ quits (~IRC@user/atum/x-2392232) (Ping timeout: 245 seconds) |
| 2021-07-29 18:15:43 | <motle> | logs stop on 21.07.29 |
| 2021-07-29 18:16:05 | <motle> | roboguy_: i never got AD, but yeah i tried a bunch |
| 2021-07-29 18:16:24 | <[exa]> | yeah that's the migration date, there should be a repo linked with older ones |
| 2021-07-29 18:16:42 | <roboguy_> | did you see those papers? I think they'd probably help |
| 2021-07-29 18:17:17 | <motle> | anyway, its topological locality, so stencil "foraging" does |
| 2021-07-29 18:17:17 | <motle> | its more about the neighbour connectivity as per edges |
| 2021-07-29 18:17:17 | <motle> | i mean, in terms of "differentiable datatypes" (SPJ) |
| 2021-07-29 18:17:17 | <roboguy_> | they're all about implementing those things nicely in Haskell, leveraging the type system |
| 2021-07-29 18:17:27 | <motle> | you get multiple derivatives in cartesian grids |
| 2021-07-29 18:17:27 | → | curiousgay joins (~curiousga@77-120-186-48.kha.volia.net) |
| 2021-07-29 18:17:29 | <roboguy_> | differentiable datatypes are something else entirely |
| 2021-07-29 18:17:33 | <motle> | and something messed up with graphs |
| 2021-07-29 18:17:45 | <motle> | but its like a helish topological calculus or something |
| 2021-07-29 18:17:56 | <motle> | best just stick with "navigating nearby" |
| 2021-07-29 18:18:02 | <motle> | pointers is fine |
| 2021-07-29 18:18:12 | <motle> | the comonad really |
| 2021-07-29 18:18:19 | <motle> | yeah its the navigable pointer comonad |
| 2021-07-29 18:18:39 | <motle> | easyer for us than any "differentiability" |
| 2021-07-29 18:18:45 | <motle> | just to get at "local" |
| 2021-07-29 18:18:58 | <motle> | topology helps but barely |
| 2021-07-29 18:19:18 | <motle> | "neighbourhood" |
| 2021-07-29 18:19:29 | <motle> | but we just say convolutional stencil |
| 2021-07-29 18:20:02 | <motle> | so its like the convolutional capability of graph pointer comonads? |
| 2021-07-29 18:21:27 | × | fossdd quits (~fossdd@sourcehut/user/fossdd) (Ping timeout: 240 seconds) |
| 2021-07-29 18:21:27 | <motle> | roboguy_: yeah i said idk what AD at all |
| 2021-07-29 18:21:27 | <motle> | says it speeds up learning, never saw how, gibberish |
| 2021-07-29 18:21:36 | <lechner> | wow, that's like james joyce |
| 2021-07-29 18:21:38 | <motle> | was doing 0 derivative stochastic learning for ages too |
| 2021-07-29 18:21:47 | → | lavaman joins (~lavaman@98.38.249.169) |
| 2021-07-29 18:21:53 | → | fossdd joins (~fossdd@sourcehut/user/fossdd) |
| 2021-07-29 18:22:13 | <motle> | some things are just genuinely incomprehensible to me for whatever reason and AD was one of those |
| 2021-07-29 18:22:16 | <roboguy_> | motle: automatic differentiation essentially allows you to take the derivative of a function written in a programming language |
| 2021-07-29 18:22:35 | → | waleee joins (~waleee@2001:9b0:216:8200:d457:9189:7843:1dbd) |
| 2021-07-29 18:22:45 | <motle> | yeah, which makes also as much sense by tautology |
| 2021-07-29 18:22:47 | × | lavaman quits (~lavaman@98.38.249.169) (Remote host closed the connection) |
| 2021-07-29 18:22:52 | <roboguy_> | what |
| 2021-07-29 18:23:00 | → | lavaman joins (~lavaman@98.38.249.169) |
| 2021-07-29 18:23:10 | <motle> | AD is program differentiation |
| 2021-07-29 18:23:16 | <motle> | ok.... so.... |
| 2021-07-29 18:23:24 | <motle> | monads are also burritos! |
| 2021-07-29 18:23:46 | <motle> | what does it mean to differentiate a program? |
| 2021-07-29 18:24:02 | → | burnsidesLlama joins (~burnsides@dhcp168-019.wadham.ox.ac.uk) |
| 2021-07-29 18:24:04 | <roboguy_> | I think you are mixing up two things. Like I said, it is unrelated to the datatype derivative thing |
| 2021-07-29 18:24:14 | <motle> | thats what its not so... |
| 2021-07-29 18:24:23 | <motle> | what is it!? |
| 2021-07-29 18:24:27 | <motle> | idk go read about it |
| 2021-07-29 18:24:31 | <motle> | i dont care! |
| 2021-07-29 18:24:33 | <janus> | motle: do you also use other social media? i wish to subscribe to your newsletter |
| 2021-07-29 18:24:54 | <motle> | really i think AD is just sillyness |
| 2021-07-29 18:25:09 | <roboguy_> | For example, if you have a function `f x = 2 * sin x`, then there is a function called `diff` that allows `diff f` to be \z -> 2*cos z |
| 2021-07-29 18:25:15 | <motle> | noooo |
| 2021-07-29 18:25:18 | <motle> | duh |
| 2021-07-29 18:25:26 | <motle> | i know how to differentiate after type restriction |
| 2021-07-29 18:25:33 | <motle> | absurd |
| 2021-07-29 18:25:34 | <roboguy_> | type restriction? |
| 2021-07-29 18:25:40 | <motle> | your doubles |
| 2021-07-29 18:25:51 | <juri_> | motle: please be a bit kinder. people are trying to be helpful. :) |
| 2021-07-29 18:25:52 | <motle> | i can differentiate that no problem |
| 2021-07-29 18:25:59 | <motle> | noo, its not a real concept! |
| 2021-07-29 18:26:05 | <motle> | they are skirting this! |
| 2021-07-29 18:26:21 | <motle> | its layer upon layer of misdirection *on all the layers* |
| 2021-07-29 18:26:25 | → | Atum_ joins (~IRC@user/atum/x-2392232) |
| 2021-07-29 18:26:40 | <roboguy_> | I think I cannot help very much in this discussion, but I hope you can make progress on your project |
| 2021-07-29 18:26:48 | <motle> | "oh, you mean regular differentiation, what were you trying to get me to read about?" |
| 2021-07-29 18:27:10 | <motle> | a feux thing exists in the notion of AD and its not your fault and im sorry |
| 2021-07-29 18:27:18 | <motle> | but im pissed at AD nonetheless |
| 2021-07-29 18:27:22 | <roboguy_> | that's weird |
| 2021-07-29 18:27:26 | <motle> | yeah! |
| 2021-07-29 18:27:41 | → | Atum__ joins (~IRC@user/atum/x-2392232) |
| 2021-07-29 18:28:05 | <motle> | if it wasnt a tautological mess i wouldnt be complaining so... |
| 2021-07-29 18:28:13 | <roboguy_> | define "tautology" |
| 2021-07-29 18:28:13 | → | spirit_ joins (~spirit@171.61.156.55) |
| 2021-07-29 18:28:28 | <motle> | tautologies are like this... |
| 2021-07-29 18:28:47 | <motle> | er, wait thats actually a good definition, oh wait no its not... argh |
| 2021-07-29 18:29:08 | <motle> | idk what that is, seems badly broken |
| 2021-07-29 18:29:15 | → | echoone joins (~echoone@188.74.32.13) |
| 2021-07-29 18:29:19 | <motle> | dont make theories like that plz! |
| 2021-07-29 18:29:27 | <roboguy_> | tautologies are logical propositions of the form "p implies p" |
All times are in UTC.