Juliaの@ code_warntype便利すぎでは?
Retweeted by 黒木玄 Gen Kuroki
retweeted at 22:07:07
Stats | Twitter歴 5,099日(2010/04/13より) |
ツイート数 315,592(61.8件/日) |
表示するツイート :
Juliaの@ code_warntype便利すぎでは?
Retweeted by 黒木玄 Gen Kuroki
retweeted at 22:07:07
コードも公開されている!(学習済みパラメータは非商用での公開)
https://github.com/deepmind/alphafold…
Retweeted by 黒木玄 Gen Kuroki
retweeted at 21:42:38
SHIMANAMI Ryo / 島並良@shimanamiryo
これぞ神戸大学のウリ https://pic.twitter.com/7iwnsFzKw8
Retweeted by 黒木玄 Gen Kuroki
retweeted at 21:24:44
#Julia言語 で分岐解析するためのパッケージを見つけたので遊びたいものリストに入れた(URLが古かったので差し替え)
https://juliahub.com/docs/BifurcationKit/I1INQ/0.1.5/…
Retweeted by 黒木玄 Gen Kuroki
retweeted at 21:17:27
Dillon Niederhut PhD@dillonniederhut
New in plotly/dash:
Python 2.7 is out.
Julia is in.
- @nicolaskruchten at #SciPy2021 https://pic.twitter.com/yvT0AgksL8
Retweeted by 黒木玄 Gen Kuroki
retweeted at 20:46:43
個人的に一番好きなのはn=6の場合です。 https://pic.twitter.com/I96wsV8Kw6
Retweeted by 黒木玄 Gen Kuroki
retweeted at 20:44:26
n=5の場合も描けた。綺麗だなぁ。
Excelやればできるじゃん。 https://pic.twitter.com/LipzdPY4vd
Retweeted by 黒木玄 Gen Kuroki
retweeted at 20:44:22
カラビ・ヤウ多様体、Excelで描けた。。。
(射影したもの)
#Excelアート
#カラビヤウ多様体 https://pic.twitter.com/FezM5HXbxM
Retweeted by 黒木玄 Gen Kuroki
retweeted at 20:44:19
非公開
retweeted at xx:xx:xx
これ、非常にイイ! #Julia言語
マンデルブロ集合で並列化とGPU計算
↓
https://nbviewer.jupyter.org/gist/antimon2/dd23b1e62d5fcd08d91fd420d7ba996d/Mandelbrot.mt.cu.jl.ipynb… https://twitter.com/antimon2/status/1415991186142089220…
posted at 20:32:49
@dhotta_met 補足。マルチスレッドは何パターンか試しましたが、`Threads.@ spawn`はJulia1.3から用意されたAPIなので古いJuliaだと動かないと思いますのでご注意です
Retweeted by 黒木玄 Gen Kuroki
retweeted at 20:23:25
@dhotta_met 手元の環境でJuliav1.6.1(+CUDA.jl)で試しました(コア数12なのでJULIA_NUM_THREADS=10で)が約3倍速にはなりました。allocationやオーバーヘッドでそれ以上は速くならないとしても20スレッドで倍程度にしかならないのは謎ですね…ちなみに↓参照
https://nbviewer.jupyter.org/gist/antimon2/dd23b1e62d5fcd08d91fd420d7ba996d/Mandelbrot.mt.cu.jl.ipynb…
Retweeted by 黒木玄 Gen Kuroki
retweeted at 20:23:22
Daisuke HOTTA, 堀田大介@dhotta_met
せっかく出力変数や中間変数の配列をpreallocateしてもThreads.@ threads マクロでのスレッド並列だと大量のallocationが発生してしまうみたいです。そのせいか、20スレッド使っても単一スレッドの倍くらいの速度にしなかりません。これ、改善できないものでしょうか・・・
Retweeted by 黒木玄 Gen Kuroki
retweeted at 20:23:20
Daisuke HOTTA, 堀田大介@dhotta_met
#Julia言語 #JuliaLang の手習いにマンデルブロ集合の計算をスレッド並列とGPUで書いてみました。スレッド並列もGPUも、ものすごく手軽に書けてビックリ。こういうembarassingly parallelな計算ではGPUは本当に速く、CPU単一スレッドと比べて千倍くらい速くなりました。 https://pic.twitter.com/DcBTiXlm0M
Retweeted by 黒木玄 Gen Kuroki
retweeted at 20:23:17
@kmai53 私も将棋センターとか行くと駒落としてもらわないと勝てないのが悔しくて悔しくて、、(^_^;)
Retweeted by 黒木玄 Gen Kuroki
retweeted at 17:01:14
町内会の将棋大会で
相手の70代くらいの人、負けて怒っちゃった(^_^;)
何も言えない………。
(ただ言えるのは女子高生に負けるの悔しいんだろうな、、小学生の強さをまだ知らなかった時の私みたい笑)
Retweeted by 黒木玄 Gen Kuroki
retweeted at 17:01:08
This is a "digital poster" for #SciPy2021, and I am hanging out at the poster session now if you want to say hi!
This animation is the same as the previous, but the red line is using Euler's method, rather than a leapfrog integrator. https://pic.twitter.com/HZRplvuBhf
Retweeted by 黒木玄 Gen Kuroki
retweeted at 17:00:01
New post on step size and mass matrix adaptation for HMC/NUTS:
https://colcarroll.github.io/mcmc-adapt/ https://pic.twitter.com/a33OZl9lYd
Retweeted by 黒木玄 Gen Kuroki
retweeted at 16:59:57
その半数くらいからは Ma’am と呼ばれているが🤔
Retweeted by 黒木玄 Gen Kuroki
retweeted at 16:58:02
外勤先でワクチン打ってて英語しか話せない人の対応すると「日本語話せないのですごく困ってたんです、本当にありがとう!!」ってすごく感謝してもらえて、日本の方の診療をしていてあまりストレートに感謝されることなかったから率直に照れてしまう。
Retweeted by 黒木玄 Gen Kuroki
retweeted at 16:58:00
Don’t worry about a thing, cause every little thing is gonna be all right.. https://pic.twitter.com/VoC77LBWZf
Retweeted by 黒木玄 Gen Kuroki
retweeted at 13:14:52
Re: RTs #Julia言語 解説
以下のプルリクエストのタイトルが
optimize optimizer
となっていて、「どんだけ最適化が好きなんだ!」という感じでうけている。
https://github.com/JuliaLang/julia/pull/41580…
posted at 13:13:57
Stefan Karpinski@StefanKarpinski
@ChadScherrer @quinn_jacobd So does Base Julia. But it will happen
Retweeted by 黒木玄 Gen Kuroki
retweeted at 13:10:07
非公開
retweeted at xx:xx:xx
https://github.com/JuliaLang/julia/pull/41580… https://pic.twitter.com/i1vHbsj8OA
Retweeted by 黒木玄 Gen Kuroki
retweeted at 13:09:51
Stefan Karpinski@StefanKarpinski
@tgamblin @hipsterelectron @miguelraz_ @vamchale @code_report @acidflask @pantsbuild @spackpm We build dozens of ABI variations so we can support different setups; there is an override system if you want to use a specific vendor library for something like MPI or BLAS
Retweeted by 黒木玄 Gen Kuroki
retweeted at 13:09:08
Stefan Karpinski@StefanKarpinski
@tgamblin @hipsterelectron @miguelraz_ @vamchale @code_report @acidflask @pantsbuild @spackpm I also don’t think the others have immutable installation, which is a pretty significant feature from a reproducibility perspective: the *only* install step is unpacking a tarball.
Retweeted by 黒木玄 Gen Kuroki
retweeted at 13:08:08
Stefan Karpinski@StefanKarpinski
@tgamblin @hipsterelectron @miguelraz_ @vamchale @code_report @acidflask @pantsbuild @spackpm It differs from Apt and Nix in that it is cross-platform and supports Linux, FreeBSD, MacOS and Windows. It differs from Conda by using the same tool chain to build packages for all platforms, which makes the results more similar and compatible (also easier to dev & maintain)
Retweeted by 黒木玄 Gen Kuroki
retweeted at 13:07:52
Stefan Karpinski@StefanKarpinski
@hipsterelectron @miguelraz_ @vamchale @code_report @acidflask @pantsbuild @spackpm All this is tricky to make work, but we’ve got all the tooling and it works beautifully — extremely reliable and fast. Other languages could use it too; the pre-built binary tarballs are language agnostic: http://binarybuilder.org
Retweeted by 黒木玄 Gen Kuroki
retweeted at 13:07:18
Stefan Karpinski@StefanKarpinski
@hipsterelectron @miguelraz_ @vamchale @code_report @acidflask @pantsbuild @spackpm Instead, we take as our principle that software should be immutable: we install all packages and libraries by unpacking tarballs which are then never touched, so we can perfectly reproduce that forever
Retweeted by 黒木玄 Gen Kuroki
retweeted at 13:07:03
Stefan Karpinski@StefanKarpinski
@hipsterelectron @miguelraz_ @vamchale @code_report @acidflask @pantsbuild @spackpm The (opinionated) perspective of Julia’s package manager is that if you’re building code on user machines at all, you’ve already lost the usability battle — it’s slow and it never works reliably
Retweeted by 黒木玄 Gen Kuroki
retweeted at 13:06:35
Trent Henderson@trentlikesstats
And the Lotka-Volterra model for good measure! https://pic.twitter.com/9ElwXmGblA
Retweeted by 黒木玄 Gen Kuroki
retweeted at 09:43:48
Trent Henderson@trentlikesstats
@JuliaLanguage is my new favourite thing. I cannot overstate how incredibly well-written DifferentialEquations.jl is. Took me less than 20 minutes with no prior Julia experience to figure out the syntax and code and animate the Rössler system. Can’t wait for @JuliaConOrg! https://pic.twitter.com/usNzvJQhNe
Retweeted by 黒木玄 Gen Kuroki
retweeted at 09:43:43
Dr. Chris Rackauckas@ChrisRackauckas
Thank you for the invite! Was a nice culmination to a year's worth of #sciml, pulling the results of multiple papers together to showcase how model information improves predictions over pure #ml. Also the #julialang tooling optimization for this domain.
https://www.youtube.com/watch?v=7yPU_04Unb8… https://twitter.com/ai4pan/status/1415466648937517062…
Retweeted by 黒木玄 Gen Kuroki
retweeted at 08:40:08
AI4PAN - AI for Pandemics group centered at UQ@ai4pan
Chris's talk, 'Learning Epidemic Models That Extrapolate' now on YouTube https://youtu.be/7yPU_04Unb8 @ChrisRackauckas @ynazarathy @HKhataee @AminathShausan
Retweeted by 黒木玄 Gen Kuroki
retweeted at 08:40:04
#generativeart #JuliaLang https://pic.twitter.com/MhoEtNwrM9
Retweeted by 黒木玄 Gen Kuroki
retweeted at 08:35:17
Don’t miss our talk on new features in Gridap.jl! #OpenSource #FiniteElements package for #PDEs in #julialang https://twitter.com/juliaconorg/status/1415674320131592192…
Retweeted by 黒木玄 Gen Kuroki
retweeted at 08:35:11
...and you can download the source data in CSV files which are easy to parse for #excel #numbers as well as popular languages like #Python, R or my favorite #JuliaLang to do further processing and plotting.
Retweeted by 黒木玄 Gen Kuroki
retweeted at 08:34:26
This may be old news to a lot of you. But for anyone interested in #history, #economics, #development, the #environment and statistics, then this website is awesome: https://ourworldindata.org
It lets compare all sorts of statistics between countries. It has nice visualizations....
Retweeted by 黒木玄 Gen Kuroki
retweeted at 08:34:01
非公開
retweeted at xx:xx:xx
宮城 芥川賞・石沢麻依さん 東北大学時代のレポートに小説!?(ミヤギテレビ)
#Yahooニュース
https://news.yahoo.co.jp/articles/50025377b0e5c1df03f0e9bc8beaa0b02c4fa5d6…
posted at 00:15:21
今日の研究会の招待講演が面白かったのと、前々から使いたかったのもあって、Constant-Q TransformのプログラムをJuliaで書いてた。
このPython実装の解説記事がとても役に立った→ https://www.wizard-notes.com/entry/music-analysis/cqt-implementation… https://pic.twitter.com/tPtBRxyaS7
Retweeted by 黒木玄 Gen Kuroki
retweeted at 00:09:13
何か手を指摘されたとき「それって誰情報?」と聞くのが昔からの渡辺流。合理的。
Retweeted by 黒木玄 Gen Kuroki
retweeted at 00:03:13
渡辺明名人「(誰かから代わる手を指摘されたとき)僕けっこう最近、感想戦で『それっていい手なんですか?』って聞くことあるんですよ。要するに『それってAIの手なの?』って聞いてるんですよ。変な手言われたときに。『それってAIの手ですか?』って聞くの失礼じゃないですか。盤側に対して」
Retweeted by 黒木玄 Gen Kuroki
retweeted at 00:03:11