Mikio L. Braun

ML and AI expert

@mikiobraun

@mikiobraun Twitter Memorial

19,827 tweets · 2008–2024 · 1046 threads

2015

2015 sure brought a lot of changes for me. Let's see what 2016 has in store ;) Happy New Year for who may already eligible.
3
RT @eleventhleft: @jstanier " 🎄 He's making a database He's sorting it twice SELECT * from contacts WHERE behavior = 'nice' SQL Clause is c…
RT @fs111: Long shot: anyone here with a spare #32c3 ticket, so that I can show the awesomeness to the girlfriend? Pretty please.
So far my #hackweek project included phantomjs, Scala, some bash scripts, and vowpal wabbit. ;)
3
RT @bigdata: Just out: practical tips from @mikiobraun on Scalable Machine Learning and Complex Data Analysis at Scale https://t.co/uNH0UL7…
Replying to @andyrtd
mein Lieblingszitat war "Ich weiss nich mal was delivern auf deutsch heisst!" X-D
Replying to @jorenvs
oh I'm not against using a library, but I feel that finding the right match is often quite hard.
Deciding which library to use from incomplete specs and half read docs seems to be the new core skill. But how to teach it?
1 3
Replying to @Quesada
but WhatsApp works, twitter sortof, email and web are fine, and free offline maps and nav.
Replying to @Quesada
they also seem to be moving away from inexpensive models, which is bad, too.
Replying to @Quesada
I'm still happy in general, but there really aren't that many apps. And even if there are (like Twitter) they are years behind.
RT @AcademicsSay: ICYMI: A brief guide to department meeting discourse. via @dgalef https://t.co/J2nwBMMf4t
Media
Burned through my 3GB mobile Internet per month a few days too early. I remember times when that was technically not possible due to speed.
Replying to @fs111
I had the same things with KLM flying to Edinburgh once. They also handled it suboptimally in every conceivable way :(
As a data scientist I sometimes envied classical devs b/c they didn't have to deal with uncertainty. That was before distrib. computing ;)
3 7
Wer von meinen deutschsprachigen Followern kennt eigentlich das Dehnungs-e (wie in Soest?)
1
Paket liegt mal wieder zwei Tage unbearbeitet in Rüdersdorf rum. Seit dem Streik ist das irgendwie nich mehr dasselbe, @DHLPaket :(
Oh Twitter changed some UI element. Luckily on WP I can enjoy the old ways for another 1-2 years ;)
1
RT @nickjshearer: I hadn't seen a BIOS boot like this in so long I legitimately thought it was an ad for American Megatrends at first https…
RT @jessitron: "Chasing meaning is better for your health than trying to avoid discomfort." ^
Replying to @noelwelsh
yeah, stigma's probably not that important after all... If you improve the wheel that is... ;)
Replying to @noelwelsh
agreed. But somehow, the stigma to "not reinvent the wheel" is stronger than ever.
Why is a hotdog better than eternal bliss? Nothing is better than eternal bliss and a hotdog is better than nothing.
2 5
RT @outlandishjosh: Three hard things in Distributed Systems: 2. exactly-once delivery 1. in-order messaging 2. exactly-once delivery
@treycausey in a way that's very frequentist test theory-y, explaining what it is by rejecting alternative interpretations.
@treycausey every time I check I also learn there is yet another way to define what a confidence interval is supposed to be.
I've self-labeled myself as a data scientist for so long, I almost forgot I'm a trained computer scientist.
1 4
I'm still waiting for the "5 reasons why listicles are the end of internet journalism" article.
2 4
RT @FlinkForward: WOW! Thank you for two amazing days full of inspiring talks & trainings and a fantastic #ApacheFlink community. #ff15 htt…
RT @suneelmarthi: Procedural programming vs Dataflow by @mikiobraun at Kesselhaus #FF15 http://t.co/9m4wg5kjij
Media
"My area of study has become so complex I can no longer just wikipedia the answer" 😂 @chillax7
Media
2
Replying to @noelwelsh
IMHO putting all the actual code into implicit operations to be "factored" in at compile time is my fav anti-pattern right now.
Replying to @mdreid
this was a terribly shortened account of the actual negotiations of course. Doesn't mention build systems, too, for example.
1
Replying to @noelwelsh
does it have (i) monoids, (ii) Unicode operator dsls, (iii) all of the above? ;)
RT @FlinkForward: Finally, the first Flink Forward Conference kicks off today. Registration starts at 8 am at Palais. Don't miss the delici…
"Shall we do the next project in Scala?" - "Yeah, but no scalaz, no Unicode symbols" - "Deal!" #scalaprojects
1 2
Nothing like joking over lunch how offline stores will disrupt ecommerce. "You can just buy stuff and take it home!" - "But will it scale??"
3 2
Replying to @miguno
No problems, I can redownload 300MB worth of jar files over my phone, right?
Replying to @miguno
Luckily you have to do that only once... Unless your local repo gets corrupted and they tell you to delete it and try again... ;)
RT @robertoglezcano: How people in #science see each other. #phdchat #phdlife #professor #research #phd http://t.co/VFhqSnrh05
Media
Packed room for our Data Science workshop together with the guys from idalab at #data2day.
Media
2
And #data2day's conference part is done. Flawless organization, interesting talks and great (German) atmosphere. Tomorrow: workshops.
1
New point of view from the balcony for @mesirii's closing talk on graph processing #data2day
Media
RT @AcademicsSay: I'm not procrastinating. I'm actively engaging in the disruption of traditional academic narratives via social media.
There's something utterly enjoyable about down to earth Germans talk about Big Data. #nohype #data2day
1
I've also been asked to fill in for a talk that got cancelled, so preparing some slides on scalable ML now. #data2day #lastminute
1 2
RT @SebastienBubeck: "theory is the first order term in the Taylor series of practice", Tom Cover
Back at the university to attend talk about student project done at Zalando. Trip down memory lane ;)
Replying to @PatrickMcFadin
@evolvable @fommil just like I heard people say most design patterns don't even exist in Lisp ;)
Replying to @pavlobaron
yeah. Probably depends on the competitors, but I also think it's important to not suck than to be really great. ;)
Replying to @pavlobaron
and we haven't even talked about the quality of the product yet 😅
1
OH: "Getting your product out there is more about spending less money on marketing than the money you make in return."
RT @slyphon: Fuck programming puzzles. I want to know if an interview candidate can file a bug report worth a shit.
Every time someone talks about "the monolith" I'm thinking of the opening sequence of 2001. Including apes and all.
Replying to @bigsur0
@r6q yeah. This general trend to "cut out" lambdas plus their closure is cool but also highly I transparent.
We definitely need more books which push the word pun envelope like "Awesomeness in ReST". And a new protocal called "Transit". ;)
1
Replying to @dwf
@DRMacIver yeah right, I also tend to use the different protocol levels, 4G, LTE, or even (gasp) EDGE... . ;)
Btw, what's the spoken name for mobile Internet. Anything as catchy as "wifi" in widespread use I'm unaware of?
1
RT @McFunkypants: Coders are always told "don't reinvent the wheel" and to this I respond: http://t.co/GURhLSqNQe (via @davidjbland)
Media
RT @neil_conway: Today's cloud software: so many ad hoc systems, so few reusable components.
From where I sit I can see all three sides of windows which almost gives me something like panoramic view. Very nice.
1
@blokhstein well, not to sound too dark or something, but I think it's just a fact of life.
Discussing your own mortality with your children is utterly depressing ☹ #oneofthosemornings
1 4
Okay, let us talk about Windows 10. Have you seen the feedback app? And hordes of people who essentially say "why isn't his like 8.1 😭😭😭??"
Also, somehow all those discussions about the newest frameworks and tools seem devoid of context. And I know I have cared a lot so far.
1
Five weeks into working at @ZalandoTech, what really intrigues me is seeing systems built at this scale, not real world data science.
2 5
People joking about the iPencil on Twitter. Is the Apple event already live? Actually I don't care. TWITTER IS THE TRUTH!
2
RT @ZalandoTech: .@ZalandoTech Delivery Lead @mikiobraun tops @DataSciNews' list of most-read #datascience articles for 2015: https://t.co/…
RT @slyphon: If you don't feel like a bit of an impostor, you're not picking the right projects.
Replying to @chrshmmmr
hard to say. But there is definitely a lot of Technology involved. R/Python, Hadoop/Spark, up to D3.js eg
1
Replying to @Nico
musste nur an den Schlagzeuger aus meiner Band denken, der hat gerade in nem Kila und Fhain angefangen und würde passen ;)
1
Wow, that "frontend ninja" cost me about ten followers. That's more than math jokes usually manage to do. ;)
3
"What is lodash?" Me: "Just another JavaScript library they chose to use in this project."
1
Replying to @noelwelsh
they went pretty framework-happy on that project. Hard to understand all the pieces. ;)
RT @chrisemoody: @xamat's amazingly complete deck with 200+ slides of every recommender technique tried at Netflix with pros & cons (http:/…
Replying to @superglaze
probably not. But Amazon being the global company that it is you'll interact with US people a lot.
RT @sscdotopen: how do I find the IPs and geolocation that 41M domains had in 2012? #followerpower
RT @dataScienceRet: As a dscientist, what are things I can learn from full stack developers so I can build interesting web applications? ht…
More and more my laptop feels like some localized container for virtual appliances. And not the other way round.
2 3
Very interesting intro to STUPS by Zalando's @try_except_. Week 2 really gets started on the technical infrastructure. :D
2
Replying to @fs111
It's not my place to judge whether it makes sense for them to restructure. But something about that story is so April-1st-ish... ;)
So Alphabet is real? I have to admit I check the date twice to make sure its not April 1. ;)
1 5
Replying to @mdreid
he already figured out that eventually it converges to one. And I'm trying to remember the graphical proof for that.
My 5yo son showed me the sqrt button in his calc and said enthusiastically "this one is for making a different number!!" ;)
1 10
Replying to @DRMacIver
mass education is just inherently difficult I guess. One on one tutoring would probably help.
Replying to @DRMacIver
... But going beyond that with standard textbooks never worked out for me
Replying to @DRMacIver
just reminds me how I found it always hard to properly learn Japanese. I have what I got through being raised bilingual...
Replying to @DRMacIver
totally know that feeling. Working out stuff yourself sometimes helps.
RT @maebert: What startups are made of vs. what academics think startups are made of. http://t.co/UJNlWHV4dv
Media
Just completed the first week of onboarding. Impressive amount of work they put into making this smooth and getting everyone, well, on board
Replying to @lojikil
yeah. But unless someone like GMail starts to push it I don't see how this could ever change...
Replying to @lojikil
yeah starting with format (text+attachments only, no recursion) and only direct delivery over Internet. And only Unicode!
1 1
I think in the past four days I heard more ideas about how to organize people than in the past ten years in academia. And I like it. ;)
1 3
Replying to @lojikil
totally agree. Email as a whole should be stripped down to a workable minimal set of features.
1 1
@treycausey Thanks, I'll keep that in mind ;) Although right now everything else is already blocked by the onboarding...
2
Day four and my lunch slot is already filling up with meetings. Also, my inbox started out with about 70 emails.
3
Replying to @twiecki
yeah. I'll be tech lead for search and recommendation, so it nicely fits into stuff I did so far.
Well I guess one good thing about starting a job right after vacations is that people don't wait to be entertained by your vacation stories.
6
Aaaaand there come the vultures... At least wait till I'm out of the building, will ya? #smh
2
For what it's worth, the people have never been the reason for me leaving. Thanks for these great years, @TUBerlin!
Last day at work. Farewell's have been said and gifts have been presented. And there is still work left to do. #byebyeacademia
4
Three more days to go before leaving. Been a pretty busy past months. I sorta regret I didn't find time to blog some closing thoughts.
2
Is customer support by mechanical turk a thing? Is it already happening? Sometimes it surely feels like it.
Ok, just in case you were following, key has been acting weird for some time up and removing cap/cleaning didn't work. But it finally did!
Replying to @memming
It was still barely functional. Took me ten attempts to get a character out. Already too much to break my programming flow.
Replying to @fhuszar
hehe. Luckily some key dislodging and air spray helped to restore functionality... -_-
RT @AcademicsSay: I don't suffer from overthinking, I enjoy it. Depending on how you define enjoy. And overthinking.
RT @sgouws: "NLP is kind of like a rabbit in the headlights of the deep learning machine, waiting to be flattened" #ICML2015 http://t.co/Pk…
Sometimes, waiting for the compiler to finish is like driving on the Autobahn. You'll get there, but it takes time.
RT @fujikanaeda: I'm hearing the "Deep Learning Conspiracy" throw-down live at #icml2015 . Well, pretty tame, but got Bengio out of his se…
RT @fhuszar: Juergen Schmidhuber asks his first question at #icml2015, commenting on originality of Yoshua Bengio goup's work. This is Spar…
Replying to @notsamandjetsam
my wife is a trained mathematician, but I guess she thinks I'm just confusing my daughter ;)
Both my wife and daughter disapprove of my repeated assertion that arithmetic becomes really interesting only in quotient rings.
1 2
Massive heat wave in DE at 34C. I should point out that air conditioners aren't commonplace here.
2
Replying to @aCraigPfeifer
Ah, no I really meant that I need a faster computer because coding is so slow because the editor isn't fast enough ;)
Replying to @aCraigPfeifer
yes, let's hope it doesn't come to that (and I meant SSDs, of course..)
I definitely need a faster machine to do Big Data development. Mostly because my IDE is too sluggish given the size of Flink. ;)
2
Big project meeting 2 of 2 today. And probably the last one in an academic context. #itistime
3
This implied analogy between leaving academia and passing away is really making me uncomfortable. ;) Gotta get my stuff sorted out.
1
RT @sethjuarez: Just recorded my first episode for an upcoming data science type podcast with @mikiobraun - look for it in the coming weeks!
@zenogantner *cough* Soooo, will Big Data go the way of JavaScript, saying hey, you can do X with Big Data, too? ;)
1
Sorry for spamming, but this is really nice. Flink + Zeppelin = interactive graphs as in word counts in the Bible.
Media
4 5
Replying to @mfcabrera
@ApacheFlink actually there is, but you still need to code in the dataflow model.
Now @stsffap is doing an interactive demo of data analysis on Flink. With the new Zeppelin notebook!
Media
1 3
Replying to @noelwelsh
actually I think what would help is if a function could also return additional new implicits ;)
Replying to @derveloper
@DHLPaket hier ähnlich. Eine Woche im Lager, dann angeblich keiner da. Dann nochmal zugestellt, Abbruch wegen falscher Beladung.
Replying to @noelwelsh
slowly getting the hang of it. ;) Pretty big departure from numpy style coding. I'll cover that tomorrow at the Flink meetup.
Replying to @noelwelsh
in this specific case TypeInformation tags in Flink that you seem to require even for all intermediate results.
Scala's implicit bleeding into everything once you use them is seriously worse than C++'s const back in the days ;)
4
One day I want to learn the story about why the Berkeley and Stanford stats and ML people have no major involvement in Spark.
2 5
RT @rwhitcomb: @fs111 @mikiobraun I believe there are some snippets of xml in poms that were written once years ago and cargo-culted b/t al…
Replying to @agibsonccc
@akm I think this is even a somewhat legitimate question as HDF5 is used to store scientific data.
1
Now I can report I discussed HDF5 on HDFS potential at least once seriously. Wonder about HDF5's splitability, though.
1 2
There is stuff which nicely reduces to a few things you need to remember once you understood it. And then there is maven.
1
The longer I think about it, the more I come to the conclusion that I don't like stuff which doesn't get simpler once you understood it.
1 2
At the end of this semester I will have churned out about 500 slides at a steady rate of 45 per week. 😥
1 2
Great chance meeting with deeplearning4j's @agibsonccc at @dataArtisans talking about all things numerical Java.
5 8
RT @karpathy: #cvpr2015 one does not simply go to the deep learning workshop http://t.co/ScyXqfLjat
Media
So finally I declare defeat against tersely written notationally inconsistent sparse GP papers for today's lecture :(
1
Sometimes I wonder what is worse, Ruby's magic let-me-hook-into-variable-not-found or Scala's implicits. At least they're typed... .
I've started to contributing to @ApacheFlink's ML efforts. Going from a two person project to such a big OSS project is... interesting. ;)
4 15
[de] Mir scheint die @db_bahn gibt sich besonders viel Mühe heute mit mir. Verspätung, Zugausfall und jetzt schleichen wir wieder. :(
What it takes to become a good programmer is to go back to code you remember being proud of and not understanding a single thing.
4 7
Replying to @rmetzger_
as someone said, the hard problems in computer science are cache invalidation, naming things, and logging. ;)
1 2
Ok, I wasn't aware the when you quit your job it seems customary to drop by and express disbelief that you really want to change jobs.
3
Replying to @mdreid
sadly no. It's a bit far off. And I'll start a new job on Aug 1. Don't know if they let me go on a >4 day trip right then ;)
1
RT @StephanEwen: Will be taking about @ApacheFlink at #bbuzz at 12.20h. Join if you like to hear about batch&steaming data processing with …
What, so now after bytecode is more or less stable Scala macros break compatability between minor releases?
1
Replying to @mdreid
@karpathy unfortunately Paper Deadline does not stack in the buff, only in the hit.
The biggest surprise so far about Rust is the macro system. Pattern matching over syntax trees. Reminds me of ideas we had for rhabarber.
2
Playing around with Rust. I have to say getting just a program out you can run is super refreshing after all the JVM dev.
1 2
Lecture 5/12 on scalable ML. Still takes up most of my work time. And I dreamt of subgradients last night.
6
Replying to @DRMacIver
I used to "strip -s" to see how small they can get. I think I'm ready for another straight-to-assembler language. Maybe Rust?
Ok. And then there's eventual consistency... 10,000th tweet! Thank you all for making this worthwhile!
1
Replying to @ChrisDiehl
@DataGuild fiber in Brandenburg. That might prove difficult. How about LTE?
1
Replying to @ChrisDiehl
@DataGuild yeah. I'm sure we have a spare somewhere in Brandenburg. Might need some work, though.
Grading students submission in Python, I have to say that tightly optimized numpy code hides its intention just as good as Spark code.
1 5
Replying to @noelwelsh
but yeah, Scala with implicits is a close second in terms of expressiveness ;)
Replying to @noelwelsh
Used wrong you are in dynamic language hell and have no idea what is happening anymore ;)
Replying to @noelwelsh
it's a messy language. But they really took he messiness to a whole new level. You can have hooks when variables are undefined.
Replying to @noelwelsh
hehe. No doubt about it. I just have a sweet spot in my heart for Ruby. Such a positively messy scripting language. ;)
1
Making sure only qualified people pass is different from making sure that people learn something. But the latter is harder to scale.
1 5
Replying to @noelwelsh
yeah. I guess at a univ the mindset is a lot about setting up some "barrier" to "protect" the value of the programs.
I have a confession to make. I'm dispassionate about teaching. At least the pass or fail, grade and certificate generating style of it.
1
Replying to @InkmiHq
@codemonkeyism you mean with tracking cookies? nugg.ad does this kind of predictions, I don't know about services, though.
Alright, luckily I still have Mac OS X ATLAS binaries lying around, so we continue with our path to jblas 1.2.4.
1
The final step in cutting a jblas release is always recompiling the libs under Centos 6.x to work with Linux 2.6.
Media
1
Replying to @SardelichM
hehe. Apparently that makes git autoconvert LF to CRLF and back when you work on WIndows.
1
More cross-platform fun: auto.crlf in git. Not sure that existed back when I started jblas in 2008 ;)
Replying to @ogrisel
I need to cover Linux 32bit/64bit 2.6/3.x, Mac OS X, Windows 32/64... . And usually some dev is required, too, to fix bugs
Replying to @ChrisDiehl
pretty sure you could have HDF5 on HDFS but also HDFS on HDF5 if you really tried ;)
Also, at which points have meetings become disjoint from other forms of interpersonal communication?
RT @noelwelsh: The master at work. @mikiobraun on scalable machine learning @strataconf http://t.co/aYDghhgE5p
Media
Replying to @twiecki
@martingoodson thanks! Looks like people have started to expect that from me ;)
RT @louisdorard: How to convince your boss you should come to #papisconnect on 21 May in Paris #machinelearning #predictive technology http…
Alright, now @sean_r_owen on random forests using Spark, and then it's time for my talk on scalable ML. #stratahadoop
Media
1 3
Still stuck in my hotel room working on my slides, but live streaming the keynote nevertheless. #thefutureishere
Media
2
Replying to @dataScienceRet
actually I've been told there has been some debate of netlib-java vs jblas and currently they plan to go with netlib :(
1
My daughter remarks that the word "work" sounds a lot like the sound a frog makes. I like the association ;)
2
Sometimes I think that early Rails lessons on sensible defaults and convention over configuration have all been forgotten.
1 1
Replying to @DRMacIver
I have that all the time. Too many different programming languages with non-canonical scoping rules.
It's like 2015 and I finally learned how to integrate my Google calendar with Thunderbird???
1
My three things post is making a comeback, this time driven by LinkedIn and Facebook and I have no idea from which post. This is a first.
Replying to @smjain
there is actually a paper which uses count min sketches to store very high dim means for k means clustering.
Replying to @louisdorard
no problem! Unfortunately, I have to give lectures on Thursday this semester... Already missing one because of Strata... ;)
RT @louisdorard: Everything there is to know about #papisconnect, Europe's 1st #machinelearning conference for decision makers http://t.co/…
Replying to @hpgrahsl
it is a boosting method for kernel learning which is easy to parallelize and very fast. Not available yet. I will let you know.
1
Can't believe how many outstanding pull requests I let accumulate on jblas. Some of these are almost a year old :(
1
RT @isomorphisms: Come on Amazon. It may be good, but not that good. http://t.co/E4WqoPiHV9
Media
Replying to @hpgrahsl
@dataArtisans still somewhat new to this. But Flink guys are local and I want to help them building a competitive ml lib.
Spent the afternoon with the @dataArtisans exploring their flink-ml subproject. Looking forward to contributing!
1 6
Replying to @roidrage
I actually had a friend who thought docker was something with physical containers...
RT @ChengSoonOng: Call for contributions: #machinelearning #OpenSourceSoftware MLOSS at ICML 2015 in France. @ahonkela @GaelVaroquaux http:…
Replying to @fs111
"Critical issues are being ironed out via testing and downstream adoption."
Replying to @snikolov
my prob theory prof used to say when all the stuff in his kid's room is evenly spread out, that's maximum entropy.
1
@gappy3000 best thing about is that there exists a general theorem which says it converges, but in reality, it makes a huge difference...
1
Replying to @kdnuggets
and so on. But I think if you take something like scikit-learn it's all pretty low level right now
Replying to @kdnuggets
I mean there is a lot of practical experience which would need to be put into code, trying out different feature representations
1
Replying to @kdnuggets
actually I think if you restrict it sufficiently in terms of problem domain, much more automation is possible.
Bulk reading the Amazon ML, Microsoft Azure, Google's Prediciton API docs to prepare an articel for InfoQ. #brainoverheats
4
As I had expected, putting together a new lecture from scratch is extremely time-consuming.
3
This semester I'm giving my first lecture series on scalable ML. And I just finished the first set of slides. #whew
5
Replying to @alansaid
@ChengSoonOng Hm... I remember that we had this issue before. I'll look into it... .
RT @monty_ashley: If you're an editorial cartoonist in search of a hot take on Apple Watches, 1971 Dave Berg has you covered. http://t.co/S…
Nothing says back to school like being woken up by your son having nightmares at 5am. 😫
1
And yet again, a fast and simple method gets bad reviews because reviewers feel that parts are known and, well, it's too simple.
4 7
RT @delphinel: The ICML Workshop on #MachineLearning #OpenSource Software (MLOSS) will be held in Lille on July 10th 2015 http://t.co/Y0ke0…
RT @papisdotio: Due to popular request the CfP for #papis2015 is now until the 3rd of April. Looking forward to your proposals! http://t.co…
Whoa. Someone put my latest post on HackerNews. Thanks for all the likes, retweets, and reads! #muchappreciated
1 6
RT @johnhugg: @mikiobraun @cra Except most of FoundationDB wasn’t open source. Nothing you could use on it’s own, really. SQL parser as pos…
Wrapping my head around Java 8 streams and seeing how that fits together with stream processing.
2
Replying to @MatthewOrlinski
for "Wackelkontakt?" I'm seeing quite different definitions when I look that word up ;)
Replying to @MatthewOrlinski
The closest English has is something like "loose contact." But "wackel" has such nice negative connotations.
Back in 1999 we took a trip from Bonn to the Saarland to see the total eclipse. So not that excited now tbh. It also took hours to get back.
My 8yo daughter looking at some slides. I say "This is also math". She says "What? You can calculate with letters?" Yup. 😍
5 14
What's up with all the X like a German guides with X ∈ { parent, party, life, ... }. Since when are Germans cool??
1
Replying to @DRMacIver
hehe. Well maybe it's comforting to know that that possibility is always there.
RT @Quesada: Confirming: I'll be speaking at cebit at a panel discussion on big data on Tuesday (14:45-15:30 h). Ping me if you are going &…
Great closing talk by @mjpt777 at #qconlondon on high performance distributed systems - and Java.
2 1
Replying to @munterluggauer
they pulled a few all nighters and managed to fix the bugs and get the app approved a day before the launch. #happyending
1
Interesting. Wooga uses individual teams per game also to do incremental exploration of new technology. #qconlondon
Hm. Okay. Randomly showing code snippets in the IDE is no alternative to having actual slides. ;)
1
Well today is definitely a travel intensive day. Four appointments, three locations. All praise public transport!
1
Replying to @triologmedia
@stackevil echt? Nicht essentielle Komponenten ohne reboot updaten? BUT THAT WOULD BE CRAZY! 😆
Replying to @triologmedia
@stackevil sowas von. Und dann beim nächsten Start "update 13737/48231" 😡
Replying to @DRMacIver
There are still so many opportunites for greatness. Homeomorphic Javascript? Holomorphic Javascript? Only time will tell.
RT @TheRealNimoy: A life is like a garden. Perfect moments can be had, but not preserved, except in memory. LLAP
RT @ConfluentInc: We wrote a practical guide that contains everything we know about working with realtime data streams & @apachekafka http:…
Replying to @nraychaudhuri
I'd say it was almost too much. But they all performed admirably!
Doing the course as a four day intensive course was maybe a bit ambitious. Probably takes a few months to get familiar with all the tech.
1
Teaching students MapReduce and Spark from the ground up in four days, including the whole Java toolchain has been quite an experience.
1 1
Fourth and final day of my Big Data Course. Today we'll do some actual data analysis ;)
1 3
Also, there are error messages and there are stack traces. And the enduser should never see the latter.
1
That Moment when you have to explain the concept of serialization in Java to explain why the Spark job crashes.
1
Just realized that all teaching at university is more about seeing how stuff works and less about how to use it well.
2 3
Replying to @eoinhurrell
yeah, I've heard that clojure does an above average job of hiding the complexities of the jvm toolchain.
Student discovers some Spark example code and asks whether he can't use that instead of MapReduce. NO YOURE NOT DONE YET WITH YOUR TRAINING.
1 3
RT @jaykreps: @strlen @medriscoll @mabb0tt Google search is also mostly just materialized view maintenance :-)
My relationship to pianos is best described as that to a rental car. Sometimes you complain but it doesn't change a thing.
The amount of pieces to track to deal with bare Hadoop for the untrained mind is apparently staggering.
RT @fs111: @mikiobraun maybe every university class should state the truth upfront: "the reality out there is a big mess. You must learn to…
Replying to @fs111
yup. Sadly, that's the truth. And dealing with projects with insufficient and/or faulty documentation.
Replying to @glowingfreezer
I think people should at least once know how to do it at the cmdline. Also maven is wonky and all but a standard.
You never know how such courses turn out. My biggest concern before it begins: Will they be able to stomach maven?
3
Will spend the next four days teaching Hadoop&friends to master students at TU Berlin. Let's see how this pans out.
2
Busy preparing a Big Data course for next week. Actually this feels more like preparing for a JVM toolchain course. #maaaaaaaaven
3
RT @mhausenblas: @mikiobraun @kingmesal @stevendborrelli @ApacheMesos @ApacheMyriad thanks & believe me I also know *exactly* what I'm talk…
Replying to @mhausenblas
@kingmesal @stevendborrelli @ApacheMesos @ApacheMyriad liked the kids birthday party analogon. #tooclosetohome ;)
1
Replying to @Quesada
Main reason I got this, however was the large battery (it's a 6" phablet), and the price, about 200€.
Replying to @Quesada
It's good. Was fed up with Android's lack of updates and overall sluggish performance in spite of good hardware.
Best reaction to my new phone: "Nokia? I thought they were acquired by Microsoft!" Just as in all acqui-hires, the brand ceases to exist. ;)
1
Actually, WP isn't that bad. App-wise it's a bit like Android three years ago, apps lagging behind compared to iOS ;)
1
So I exchanged my Galaxy S3 for a Lumia 1320. And now that WP10 is pre-released I'm like "as if I'll ever get an update." But wait... .
Replying to @cartazio
thanks. Yeah. Still not sure where this is going, but I guess there will be more changes soon... .
RT @strataconf: Data experts @slangevi @mikiobraun @datamusing @StatGarrett + more present at #StrataHadoop London #DataScience Track http:…
Replying to @IgorBrigadir
@clockworkmod I see. Yeah, good thing users can fix this easily. No need to catch that error automatically. 😈
1
Replying to @markusandrezak
true. Somehow they both need to contribute to the company goals, right?
This is pretty great. A friend's Android is broken because the root partition is mounted read only. Solution: root+fsck+remount. #easyaspi
Replying to @markusandrezak
@jboogie yeah. Why is it that intrinsically hard? I'd guess almost disjoint goals. If no one mannages the bridge, it fails
RT @debasishg: m/c learning pipelines on a distributed platform .. tweak hyperparameters and see impact all the way down.. Spark FTW http:/…
@treycausey OK, even if I did I would only fix typos and notation, never touch the data, obviously... 'o_O
@treycausey yup. Looks like at least I don't have to edit figures in inkscape by hand this time ;) #thatmustbeanoutlier
TFW you lose confidence in git's automerge capabilities and resorting to 'git fetch'/'git merge'.
1 1
Replying to @pavlobaron
yeah. If only Google weren't that picky about high rate usage of their search API we could have that this afternoon.
Replying to @pavlobaron
that's like googling for an idea you had to find out whether someone already did it and opensourced it. 😜
ICML deadline drawing near. Time to edit the LaTeX sources like a grad student ;) #actuallyenjoyingthis #sorta
1 4
RT @StephenPiment: At #FoundationDB, our CEO & CTO write software. @beaucronin “one simply does not write software at my level anymore” htt…
Before scaling out, a machine learning person would always try some approximation shortcut to achieve speed up. #cheating #orisit
3
Interesting talk by Columbia's Stephan Mandt on large scale Bayesian inference with stochastic variational inf. No parallelization, though.
1
Speaking of unfulfilled desires and such, last night I dreamt I finally got the kitkat update for my phone. #androidisbroken
1 1
When the coffee shop is apparently streaming music of a free Spotify account and you hear their commercials every ten minutes.
1 2
RT @dataScienceRet: Computing Recommendations at Extreme Scale with Apache Flink and Google Compute Engine Ht @mikiobraun http://t.co/qlf0j…
Man, GraphLab/Dato and Spark companies are certainly pushing some serious marketing $$$ into the Big Data Science hype.
TFW you take care of last minute formalities to make sure a project actually gets funded. #finallysomeimpact #millionsperminute ;)
What always makes me laugh out loud is when the hadoop command says "couldn't load class <mistyped arg>". WTF ARE YOU TALKING ABOUT???
I also continued with my quest to set up Hadoop from the bare OSS distros. Let's put it this way: a polished UX feels different.
These Big Data benchmarks are getting out of hand. Data analysis being such a diverse field, more than enough opportunities to "cheat".
3
Wait, Always On + Xbox one + Windows 10 = remote mobile gaming. But nooooo, we bitched about Always On and now it's gone.
Replying to @davidandrzej
I understand where it comes from. It's just a bit counter intuitive that the estimate doesn't become better over time.
Hm. The theoretical guarantees of HyperLogLog depend only on the actual # distinct elem and the # bins, but not the # events?
1 2
Replying to @noelwelsh
yeah, you can talk about these things informally, too, but sometimes a formula says more than 1000 words. ;)
Well, taking the practical approach to data science is nice and all, but talking with someone who knows linear algebra is nice, too.
1 6
RT @BWJones: I’m answering Tweets at this point. Not answering emails until after this grant goes in. email = more tasks. Twitter = Dopam…
RT @jaykreps: The dirty secret of non-trivial iterative algos on "big data frameworks": result is often slower than a single thread http://…
Replying to @mdreid
I was also thinking something like rate at which sticks disappear based on how much you travel, or how many colleagues you have.
1
Replying to @dosinga
hehe, yeah that's right. One of my scissors finally turned up from under the couch the other day.
1
Replying to @kostas_tzoumas
very impressive progress! BTW, do you think you can add an RSS feed to the flink blog?
I wonder how many USB sticks you need to buy till equilibrium is achieved and you always find one wherever you are.
4 3
RT @thegrugq: Australia is f’king hardcore even the koalas (poisonous!) wear boxing gloves. http://t.co/FkU6qxJKMi
Media
RT @michameder: now listening to @kostas_tzoumas talking about @ApacheFlink @telefonicaid together with @matip http://t.co/s2mh3fzVif
Media
Considering machine learning used to be some advanced course avoided by most due to large amounts of math, this is all pretty remarkable.
1 3
RT @mrphlip: @karinjiri @freebsdgirl enum ಠ_ಠ {°□°╰, °Д°╰, ಠ益ಠ╰}; void ┻━┻︵​╰(ಠ_ಠ level) {exit((int)level + 1);} if (error) ┻━┻︵​╰(°□°…
What?? "Chinese TV giant TCL brings Palm Inc. back to life" http://t.co/p9byflVGW6 (via @engadget)
Oh cool, Autodesk Sketchbook, the app with which I make the figures for my talks, has a new version out.
RT @genetics_blog: no no no no no RT @GaryGJohnson: vim users get ready to accidentally power down your new Macbook every 5 seconds http://…
You're probably thinking why am I bitching about reports but they're one of the few formal requirements linked to millions of funding money.
RT @DRMacIver: @mikiobraun Pssht, "off-topic". I don't even know what on topic would be on twitter. :-)
Replying to @DRMacIver
hm. apparently I can't DM back because you're not following me (which is totally ok, of course). Just letting you know ;)
Replying to @DRMacIver
Yeah, I can imagine. Hopefully your pain will help others to not fall for the same traps.
Feeling a bit like Jack Nicholson in A Few Good Men. I'm the guy who does the dirty work so that others can live in the ivory tower ;)
A PhD students just thumbed the final report I'm working on. Never saw someone leave my room faster.
1
Replying to @DRMacIver
don't know how many doubly linked lists I've written over the years. Or matrix libraries...