Mikio L. Braun

ML and AI expert

@mikiobraun

@mikiobraun Twitter Memorial

19,827 tweets · 2008–2024 · 1046 threads

2011

Replying to @DRMacIver
I'm still using delicious. Same handle as on twitter. Can't tell whether you'd find it interesting, though ;)
RT @jaimeohm: Coders are special. "We are expected to know how to do things we've never done before and estimate how long they will take ...
Gowalla's team to join Facebook? And I only found out about it now? Well that says it all...
Replying to @owehrens
warum wundert mich das nicht... die erzählen einem doch was man hören mochte... ;-)
Alright, time to wake up my son and join the rest of my family in the church. Merry Christmas, everyone!
Replying to @Nico
Ich halte das auf tumblr natürlich für einen Witz. So sind'se halt, die tumblerers.
Replying to @Nico
ah ja. Auf tumblr geht das Gerücht um, man würde die Mütze nur bekommen, wenn man einen tumblr gold account hätte, daher die Frage...
Still amazed at how well cyanogenmod works on my old HTC Desire. So, HTC, was it really so hard to get Gingerbread on that device after all?
@Nico Die Weihnachtsmütze auf dem Blogbanner, ist das auch so eine "tumblr gold account"-Sache? ;)
I have to admit that I was blissfully unaware of that SOPA bill until someone mentioned "SOPA vapors" today.. Now this is serious, it seems.
War klar: in dem Jahr wo ich Ende November schon Winterreifen drauf und einen Klappspaten an Bord habe wird's nix mit weißer Weihnacht.
Somehow, it took the usual 3-4 days to recover from #NIPS2011. Always thought it's the jetlag, but it's probably just the information intake
1 1
RT @mdreid: I think the good folk at Twitter misnamed the "Ignore" tab by calling it "Discover" instead.
Listening to "James Taylor at Christmas". High time to get into the right Holiday mood ;) .F
Just handed my xperia mini pro to get the keyboard fixed. The "," was getting activated the whole time on neighboring keys... .
RT @Balkonschlaefer: +++ EIL +++ Diesjährige Weihnachtsansprache wird Wulffs Anwalt halten +++ EIL +++
"What? Another batch of papers? But I already reviewed those". But it was another conference, only web site looked the same. ;)
Just remembered that I was reviewing for three different conferences all using Micrsoft's CMT, all looking almost the same.
Replying to @fhuszar
Yes, you're right. It was pretty scary. About a third of the people seemed to have it at some point in time.
Iberia was quite a nuisance, though. Every flight was late, no free food, pilot strike, and my luggage didn't make it back.
1
@albert_swart yes, it looks like. Interest in the ML community seems to be very large. Still, a lot of terminology to process.
Hm... Laptop didn't power down when I closed it and put it in my backpack. Good thing I took it out before it overheated... .
Will probably still take some time till copula terminology becomes mainstream in machine learning... Who are Gumbel and Frank? #NIPS2011
1
Learning about copulas in Nicole Krämer's copula tutorial. Interesting. Workshop filled to capacity. #NIPS2011
2
If you've missed our @twimpact demo at #NIPS2011, just talk to me, I still have everything. I'll mostly be in the BigLearning workshop.
1
Finally arrived in Granada for #NIPS2011. Just to learn about the impending strike by Iberia pilots on Sunday :(
Sorta heartwarming seeing US researcher's taking double digit hours to travel to Granada. That's what we're Europeans are doing every year!
Lack of updates on #NIPS2011 tutorials... . Probably means people are actually listening ;)
1
Replying to @mkhq
I put the shortcuts for "work" workspaces and the "fun" workspaces somewhere else. Work: Alt + 1/2/3 Fun: Alt + M... No real help..
Replying to @mdreid
I'll arrive on Wednesday. Looking forward to finally meeting you in person ;)
Replying to @_onionesque
@ChrisDiehl ah, I see. Yes, I agree. Although now that you mention it, some would say you should also write it as two words...
Replying to @_onionesque
@ChrisDiehl Actually, it conists of two words, kaputt = broken, and the past tense of spielen = play. => "Played broken"
Replying to @ChrisDiehl
;) Actually, you can combine other verbs with "kaputt", e.g. "kaputtgefahren" (driven) or "kaputtgeschrieben" (written).
RT @horax: wieso lese ich in der neuen #twitter app eure vor- und zunamen. die kenne ich doch gar nicht. wo läßt sich das abstellen?
"kaputtgespielt" (German, "tinkered with till broken"): what Google is doing overhauling most of its products right now.
1 2
Running three screens on two servers as two different users side by side in a terminal can be a bit confusing... .
Replying to @DRMacIver
ah sorry, I seem to get easily confused this morning.. I thought you meant other people's replies (which I wouldn't be seeing..)
Replying to @DRMacIver
hm I wonder how Twitter's awesome conversation data model will let me do that... Ah yes, it doesn't..
Listening to Chuck Loeb's Plain'N'Simple. Very nice, relaxed. Organ-drums-guitar trio plus some extras.
1
Scala pro tip: If you extend from Iterable, make sure to implement size if there is a faster way than iterating and counting. #facepalm
Opening one of our analysis snapshots takes way longer than it's supposed to do... Wondering what's happening... .
Replying to @jowede
Bin gerade über den Vortrag gestolpert. @twimpact gibt's übrigens wirklich schon, wir machen real-time social media trending usw...
Just read that my smartphone will supposedly get ice cream sandwich in March. I won't fall for that again. Gingerbread's just fine... . #sob
Sorry, I meant to say CTRL+Tab. Three complete losses of keyboard input in 10 minutes. What is going on?
The hard disk to hold our @twimpact analysis data for the #nips2011 demo finally arrived. Now we need to copy everything over USB... 2.0..
Replying to @tirsen
actually, I'm not even sure if the argument with cyclic references is valid. Python uses ref counting and seems to do fine...
Started to write a blog post discussing Scala language features and classifying as essential or "use with care". Pretty complex stuff...
Second stage of our @twimpact analysis for #NIPS2011 demo took only about 1 day one the cluster. Now we have to wait for more filespace...
@muratk3n Or first prove that some things are invariant to crd. chngs only to show how some other things get much easier in the right crds.
RT @twimpact: Some numbers on our NIPS demo - So here are is a bit of background information on the data processing we’re... http://t.co ...
Replying to @tirsen
I'm using favorites more as a personal bookmarking, and RT to pass it on. I'm often doing both.
So if they process 120000 tweets per second using 936 cores, that makes about 128 tweets per second per core... .
Es scheint, @1und1 hat es geschafft, meine Abschlussabrechnung EINFACH SO ZU MACHEN. Wahnsinn. Endlich läuft's >;)
Just as @thinkberg has advised for a long time, I'm finally using long descriptive variable names. At least in method parameters ;)
RT @mdreid: I'm really into meta-ironic faux-intellectual post-hipsterism at the moment. You've probably never heard of it. #hyphenstoo
It's just 11:45 and the sun is already going down behind the building on the other side of the street. #sigh
Actually, I think there is some nice interaction between the camera technology Sony has and what they put into their smartphones.
Just updated my xperia mini pro. Suprisingly, there are a few nice features: swype-like keyboard, sweep panoramas, screenshots.
Replying to @mlsec
Not quite, I'm not crazy. But it is an interesting comparison of ngrams vs. dictionaries on short strings with little statistics.
A few bug fixes later much better agreement between the chrome language detector and our dictionary based on. #thisiswhyIalwaystest
As @thinkberg suggested, running some stats on chrome language detector vs. our dictionary based one. Amazing lack of agreement.
@muratk3n No problem ;) The CLD works fine, but also has it's share of false positives. Twitter speak is weird, I guess.
@muratk3n So far, we had something dictionary based, which also more or less worked. It's hard. Tweets are so short. Not enough statistics.
Looks like we finally found a reasonable way to bound the growth of the graph of users who have retweeted a tweet... Very good... .
1
Replying to @DRMacIver
I've switched to xubuntu recently. That involved deleting all of .gconf, .gnome, .gnome2 and a bunch of other dirs...
Not sure whether I like the new Google analytics. Whatever happened to the dashboard where you could quickly compare all your accounts?
The gaming aspect of Klout certainly means sense. But changing the score every other week kinda spols it.
It seems my ability to not select the Scala compile server but the actual program in jvisualvm has vastly improved recently. #yay
The problem is, of course, that you never know whether some library (e.g. a JSON parser) used substring() to extract a string or not...
Otherwise you will retain the whole string in memory, because substring is just a view of the longer string.
Today I learned the importance of saying "new String(x.substring(a,b))" if b - a is much smaller than the length of x.
1
Man, this Oliver Sorin spam is seriously getting out of hand on Google+... Or is this for real?
Alright, about 80 minutes to find the memory leak before I have to catch a plane... #crunchtime
@muratk3n But it's definitely one of those things which should've been fixed a long time ago.
@muratk3n Yes, maven is a mess. Unfortunately almost every project supports it. And once you've got it running, it's sorta OK.
This made rounds on G+ in German: "G+ is like a gym. Everyone is a member and you definitely should go there again some time."
1 1
Wednesday's revelation: my docking station doesn't work properly without the power cord plugged in.
Yay, Gregoire's paper with Klaus and me got accepted to the Deep Learning Workshop at #NIPS2011 as a poster!
Shsh, no luck with IntelliJ... Especially after cut&paste it goes into some strange loose-keystrokes-at-random mode with high probability.
Thunderbird 7 is really having bugs in all of its list displays. Added mailing list in addressbook, got two, deleted one, deleted both.
Before thinking about how to scale out your data store, ask yourself whether you really need all the data you're saving.
My email inbox is a mess... @thinkberg showed me saved search folders in Thunderbird. Let's see if that is going to help a bit.
Replying to @atpassos_ml
@mdreid @ogrisel Yes, I'm in for it, too. Whichever hashtag we'll settle on at the end ;)
Reverting back to Thunderbird 3.1... Evolution just times out on our institutes email servers :(
Something is quite wrong with Thunderbird. Messages are sorted into threads in the wrong order and recipient from fields are all mixed up :(
Replying to @DRMacIver
but all in all it's about the same level of entertaining non-sense you get everywhere. So even those elite G+ers love cat gifs.
I'm sorta relieved that the "hot on Google+" stuff is mostly cat gifs. Like everything else on the Internet.
1
Replying to @dbanes
I know what you mean. It's also extremely refreshing to see a good ol' distributed Internet architecture for a service for a change.
Replying to @agnoster
Yes, indeed... It's also going to get a lot colder before it gets warmer again ;)
Replying to @dbanes
Been reading "XMPP: The definitive guide" over the weekend. You were right, very interesting stuff!
2 1
Hm... memory leak points into direction of akka.dispatch.MessageInvocation... More digging seems necessary... .
Ah, sorry about that. Seems you can. Probably my machine was just a bit overloaded ;) #sorry
Ist mir echt unklar, wieso die es bei @1und1 nicht schaffen, eine Kündigung mit fast einem Jahr Vorlauf ordentlich abzuschließen...
Gerade hoffentlich mein endgültiges letztes Gespräch mit der @1und1 Hotline geführt. Bin froh, dass ich von dem Saftladen weg bin. #mannmann
Thunderbird in ubuntu 11.10 is really pretty buggy. Messages disappear, sender in overview is wrong... why?
Ups, another master thesis to read till tomorrow. Years of training have prepared me for that. Digest thesis, spill out recommendations.
Replying to @mkhq
Actually, the question for me is to whether implement that myself for some homebrewn data store.
Man, scala's Java collection conversion facilities let you shoot yourself in the foot in exciting and surprising ways!
Mouse pad arbitrarily stops working in Ubuntu 11.10. It's probably somehow linked to the driver for my rt2860 WLAN chips...
Driver support for my LG netbook actually got a lot worse in Ubuntu 11.10. Suspend-to-RAM lead to kernel panic after resume...
Resizing partitions so I can install MS Office for project proposal presentations. This is wrong on so many levels.
1
It's gotten so cold that I'd rather operate my smartphone with the tip of my nose than taking of my gloves...
I might be wrong, but it looks like you have no control over the system font anymore in Ubuntu 11.10...
It seems like they finally fixed xpdf in the latest version of ubuntu. Finally, unobtrusive TeXing in emacs again ;)
Replying to @sscdotopen
yes, I will. It's sort of a closed group related meeting right now anyway... .
RT @DEVOPS_BORAT: Fuckings! I tell Siri deploy Hadoop cluster and she order Oracle Big Data Appliance!
It's pretty amazing that you need to add some udev rules by hand to to take screenshots of an Android phone under Linux.
Replying to @leonpalafox
Oh yes, in particular when your design is as generic as Apple's. A rectangle with a glass front? Centered screen? Buttons? cmon
Replying to @leonpalafox
I agree that they sorta look similar, but to me, an iPad and an Android tablet are a completely different product.
The new @delicious isn't that bad. People should stop complaining about the changes and be glad that it got bought instead of being sunset.
RT @sgourley: What is this @Kred thing? It seems to be very much like this @Klout thing. Is measuring online influence just like buildin ...
You may question their ranking methodology, but @klout certainly does a great job at marketing their service.
Replying to @zoltanvarju
Definitely. And people are really willing to pay for whatever works marginally better than the rest ;)
RT @markusandrezak: [GER] Fefe's Blog lesen lohnt sich gerade sehr wg. dem Trojanerkram. Da haben die ja wirklich nen Schlamassel an der ...
Next: reducing time it takes to write a snapshot. Optimize, reduce data by filtering, or copy-on-write?
Replying to @zoltanvarju
I'm not really convinced that we're already where computers can understand sentiments in all but the most obvious cases... .
The only problem with advanced features is that it becomes increasingly difficult to write test cases for them. #tdd
Once you start with manifests and automagic generic code in Scala, you end up casting a lot to trick the compiler. Just like C! #Ilikeit
No if only fsc wouldn't always get confused when compiling in different directories in the shell...
The integration of the Scala compiler server (aka fsc) in IntelliJ IDEA seems to have been fixed. #yay
Alright, let's see whether I can fit that secondary index I finagled into one of my demo scripts. #whatcangowrong
Sorry, HTC, but still reading all these news on upcoming Gingerbread updates makes me sad. My new SE xperia has it out of the box... .
Ah, Storm is actually not "real-time on Hadoop" but the "Hadoop of real-time". In fact, it doesn't do storage at all.
Replying to @alexott_en
Yes, it always depends. Probably that's my point. Some people will just *always* say "why haven't you used project X?"
Sure, reimplementing stuff means code you have to maintain. But adding another project usually also adds more complexity than you need.
And you also didn't have to explain why you're reimplementing stuff all the time (instead of adding another dependency to your project).
Ah, the days where you could read about some algorithm and simply reimplement it instead of downloading another 250+ classes framework...
2
Retweet if you also have accidentally typed Java class names (like "java.util.TreeMap") into your browsers address field.
I wonder whether you can enhance a skip list to store secondary indices compactly. Hm. I should try the obvious first, probably.
Replying to @ian_soboroff
@ChrisDiehl Interesting. What exactly is it? A data set collection for microblog data?
Replying to @sscdotopen
From what I read, a data scientist is actually more some kind of computational statistician. No PhD, but also no engineering.
Replying to @argoneus
I was totally with him... until he said that PHP is his favorite programming language ;)
Given that amount of data coming from Twitter, it probably supports all kinds of hypotheses. What about reproducibility?
2
Replying to @mdreid
Finally, after all these years ;) I think you were among the first five people I followed on Twitter.
RT @twimpact: We have been accepted for the Demonstration Track at #NIPS2011. See you in Granada!
Replying to @lousylinguist
interesting link. But $5M isn't really that much money in research... IMHO they've promised more than is reasonable.
Replying to @zoltanvarju
ah yes. It definitely does. Many people would probably also say that a data scientist is more comp. stat. than ML.
Replying to @zoltanvarju
but with AI, ML, data science (and not to forget connectionism), it's more of a lateral move - at best...
Replying to @zoltanvarju
it often happens that a subfield gets so large it gets it's own name, leading to specialization which also has problems.
Replying to @sscdotopen
I'm not even sure what data science really means. And what is "unapplied machine learning"?
Replying to @chl
yes, and we used to say "we're going to create an artificial intelligence". Somehow, the fascination is lost ;)
Replying to @chl
yes, much broader in scope but also with much less commitment to solve any specific problem.
Another example: natural language processing to text mining. 'Processing' was already bad, but how is 'mining' better?
Somehow, the move from AI to ML to data science also documents an increasing loss of ambition.
8 2
First, artificial intelligence turned into machine learning. Now, we have data science. Rebranding every 15 years to keep the ideas fresh?
Replaced my old HTC Desire with a Sony Ericsson xperia mini pro two weeks ago. Very happy with my new phone: small and with a keyboard!
Replying to @mlsec
@thinkberg Es sieht wirklich schlimm aus, funktioniert aber noch. Dir Rückseite ist aber hin...
@muratk3n I find it also curious that there are disciplines like physics which has physicists and then there are all the scientists.
RT @mdreid: @mikiobraun I sort of agree, but every other type of scientist uses a computer too…
"Data scientist" is really the most stupid name ever. It's not like scientists don't usually deal with data... .
1 2
Replying to @chl
Now that I think of it, "data scientist" is a very funny term. It's not like scientists don't usually work with data.
Replying to @mdreid
;) The story was that the 5yo niece of one of my colleagues said that 1 != 2/2: one full plate is different from two half plates.
Lunch discussion's conclusion: Mathematical abstractions must not be confused with reality.
Apparently, every existing oauth library basically reimplements its own http client. If you're lucky, it does everything you need.
.@thinkberg is adding oauth support to our Twitter streaming backend. And it's about as painful as it sounds.
Thinking about ways to add secondary indices to our @twimpact backend for more efficient search...
RT @sinanaral: "@TheSocialMined: #Social #Analytics The Federal Reserve Plans To Identify 'Key Bloggers' and Monitor them http://t.co/Vm ...
Replying to @jliyi
I guess they want to know what people actually click on. Otherwise it's not that useful. It's not like they expand bit.ly URLs first.
Replying to @jliyi
Yes, that seems to be the case. At least they put the expansions for all t.co links into the tweet's metadata.
Seems like I'll be doing mostly grant proposal paperwork for the next 36 hours. #waitthisaintevenrocketscience.
@muratk3n It certainly was. Drug dealers lamenting the kind of people they have to deal with in their jobs made me really crack up.
Somehow, the percentage of Germans I follow is much more higher on Google+ than on Twitter. No idea why.
My idea turns out to uncannily tune into the actual trends when I take data where something big was happening. Other days, it's very quiet.
1
I have an idea for @twimpact which is similar to the Twitter trends, only we can compute it in a snap and also for arbitrary search queries.
Replying to @heiko
Musste gestern mit dem Rad schließlich eine Station mit der S-Bahn überbrücken, weil sonst kein Durchkommen war. :(
We're currently trying to improve the media trends at @twimpact. However, there is an incredible amount of NSFW stuff in there... #wtf
Replying to @csshsh
ah, sorry, thought you were with QuidLabs. Ok, I'd say in comparison you're still somewhat short on information... No hard feelings
Replying to @csshsh
I see. Well, I think it was well worth the wait. Very nice website design!
Just brought my car to the dealer to have it checked. Way too early in the morning. #Appointmentatseven
RT @MattTucker: We support The Patent Pledge: No first use of software patents against companies with less than 25 people. http://t.co/e ...
RT @mdreid: People paraphrasing Einstein should follow his advice and just say "Make things as simple as possible".
RT @derdino: Denke die #FDP sollte jetzt in eine geordnete Insolvenz gehen - da darf es keine Denkverbote geben
So, jetzt müssen die #Piraten mal zeigen, ob sie auch im politischen Tagesgeschäft was taugen... #ahw11
Four of our PostDocs are leaving soon and we're having the farewell party tonight. Right now, it's time for the presents...
Finally I have enough memory to run the Optimistic Weather app. Not that it does anything useful, but it's fun to watch.
I think you could sink a whole lifetime into collecting URL schemes for image posting services.
Just had the very unreal feeling that social media doesn't solve any problems at all. #scary
Working on parsing embedded images in tweets. It's a bottomless pit. Please design a reasonable API if you want your service used.
Every time someone writes "has x MB of internal memory, but can be expanded with microSD cards" about an Android phone, a robot dies.
1
Just as expected, we cannot yet upgrade to Scala 2.9.1, because akka depends on 2.9.0. What brought this #versionhell upon Scala???
Almost no reviews of Android phones report the actual amount of internal memory free for apps, just the total built-in amount.
And now Motorola is already announcing the Pro+, which supposedly will be available October. ORLY? Ah, you didn't say which year ;)
I'm really beginning to wondering about Motorola's product pipeline. The Pro has been announced forever but it's only available by Vodafone.
Instead, the media likes to make us believe it is the vision and determination of the P.I. which is sufficient for scientific progress.
To me it seems that often there is also a lot of luck involved. You set out to solve some problem, but you can't really foresee the answers.
Some examples: buildings are either run down or extremely high tech, the researcher is a visionary who is still fighting for acceptance...
Probably it's just me but every time I read an article about scientists, it seems like they're using the oldest cliches to frame the story.
Been playing around with my old Nokia E61i a bit. Incredible that we ever put up with that stuff. Access Point Groups? Seriously?
Samsung still cannot sell their tabs in most of Europe. As if a geek would mistake an Android tab for an iPad because they look similar...
1
As a test, trying to expand ALL THE LINKs. It's really amazing how much redirecting is going on even on normal sites.
Replying to @headius
Oh yes, 8G is the new 640k. It's so much, I can't imagine anyone would ever need more. Well until they do, of course.
Too late, my brain has already started to devise some simple heuristic to detect URL shorteners... . #gah
URL shorteners wouldn't be such a pain if they at least were one-to-one within each service.
Must resist urge to write some adaptive URL-shortener-detector based on URL length/mix of weird characters and domain ending in ".ly"...
Updating the link expansion code in @twimpact. The sheer amount of URL shorteners is amazing. Just because of the 140c restriction...
Scala cold start to execute a println("hello") script: v2.9.1: 3s, v2.9.0.1: 5.2s. Also REPL first shows prompts, loads libs in background.
The amount of metadata you get from Twitter in their JSON streams is really amazing. Every time I check, something else is in there.
No that would be a reason to upgrade! RT @timperrett: Just upgraded to Scala 2.9.1. Finally the REPL boots up in super quick time!
Replying to @jliyi
Yes, exactly. I've put some in Google Reader and check in every now and then.
Replying to @jliyi
Yes, they should. I like tumblr a lot, but it's hard to follow a balanced crowd without a few drowning everyone with their volume.
I could probably upgrade to Scala 2.9.1. Nah... better wait for all those libs to catch up first.
Replying to @imwithpanda
@untiefen Ich wünschte, ich könnte das von meinem Kaffee auch sagen ;)
Facebook first made quite a fuss about this. Later: "nah, it's okay, just don't make them look like a real Like button". Yeah, sure... .
Collecting some stats on the language which comes with tweets and what our language detector thinks it is.
Now only if Scala's fsc (compile server) wouldn't get confused every time I switch between console/maven/IDEA, that would be great. ;)
I've also come to like Scala's scripts as an informal way to test some ideas very much. Once it works, you can put it into unit tests.
Looking at a list of @NIPSConference Workshops. I guess there would be some interesting possibilities to present something about @twimpact.
Writing equals() in Scala for generic classes always ends in a match-and-cast fest... . #sigh
I've been sorta collecting material for that (final) post on peer review. But it's just too much... Better stop now, as long as I can... .
RT @DRMacIver: It's become so common place that I rarely notice these days, but fuck do I love having a smart phone
Replying to @martinfowler
Ah, the dangerous phase where you've stopped caring if a piece of equipment breaks because you already decided to replace it.
Apparently, when you search Google for "ruby module function" my rant on the weird scoping rules in Ruby is one of the top hits. Weird.
Replying to @ChrisDiehl
Yeah, I probably will. Already had an interesting discussion with my colleagues about it over lunch which validated my points.
Although I'm quite ok with the turn-out. Some of my colleagues have been hit much harder with bad reviews.
Let's see how much longer I can hold back on that peer review related post now that my NIPS paper is rejected ;)
Dear EA, do you really expect me to read all of the 50 pages of updated privacy policy just so I can play Dragon Age II? #wtf
Facebook, you're seriously asking me to only friend people I know in RL? That's so 2003...
1
Replying to @DRMacIver
I see ;) Just have to make sure to cancel ongoing computations while watching. Don't want laptop to overheat for nothing.
RT @ChrisDiehl: See folks all the time spending little time defining the problem. Asking the right questions is challenging. Perspective ...
Replying to @DRMacIver
Ok, you got me... . Do you mean actualy baby weasels or that cat apparently named Baby Weasel?
Hottest day of the year. Feels like you'd want to have a bucket full of ice cold water under your desk to put your feet in... .
Replying to @atpassos_ml
Yes, just like you don't use polynom division to compute the eigenvalues from the characteristic polynomial.
I guess that's why they call it a technology stack... . You just stack layer upon layer of abstraction until everything tumbles over.
Putting together small demo of trend sparklines, this time in HTML. Scala+protovis+scalate/mustache+auto-reloading epiphany browser. Nice.
Replying to @jamesgolick
oh yes. What amazes me most are those js conferences where almost all talks say "hey, we reimplemented X in JavaScript"
RT @twimpact: Last days Virginia earthquake in the hashtag cloud - Yesterday, there was a minor earthquake in Virginia at... http://t.co ...
RT @twimpact: Photo: Introducing the teamfollowback hashtag cluster. We’ve been playing around with real-time clustering... http://t.co/ ...
Just in case you're wondering #frameworkheimer - someone involved with excessive framework building.
Playing around with d3.js force graphs. Trying to get around setting up a full webapp to do that. #frameworkheimer
Actors being nice and all, what I don't like is that introducing them into sequential code for performance reasons is pretty disruptive.
Almost had that "tell me again why Twitter is useful" discussion with some colleagues again yesterday.
Up till Aug 6, they were still posting updates on the TouchPad, and then, 13 days later, they cancel everything?
Accidentally left my laptop in my office. Just noticed today when I was packing my bag. #familydad
Pretty odd, every time there is some bug in IntellJ's IDEA Scala plugin, there is already an update available for it. #causeandeffect
Hoffe mal, Mehdorn setzt bei Air Berlin nicht auch die Sparzange bei den Wartungsintervallen an.
@muratk3n Well, I guess it has the potential. I only get a bit suspicious if most people who claim this are from Berlin themselves.
Oh great, forgot to commit my changes after putting tons of docs in before starting on the next feature. #theyhavetowaitIguess
Bought Silje Nergaard's "Darkness Out of Blue" for 5€ from some bargain table a few weeks ago. It's really nice.
1
RT @markonen: You wouldn’t believe how many SQL injection vulnerabilities I keep discovering just because the name of my street has an a ...
@hgascon My favorite example are check-ins in Maps/Latitude vs. Google+. Extra confusion if you need to add the place first.
@hgascon oh yes. Some things are so complex that it takes a degree in CS and a good understanding of the structure to know what'll happen.
The confusing integration between the different Google services like Plus, Maps, Picasa, and Latitude is starting to rival that of MS Office
1
Replying to @agnoster
With horribly broken, are you referring to Android Honeycomb or the patent system?
Replying to @agnoster
I agree. I certainly don't feel sorry for Samsung or Google. But for us consumers, it means less choice. Which is a pity.
Replying to @agnoster
Yes, and that's what I find annoying about Apple. Well, I guess they are just defending their business, but still...
Replying to @agnoster
You sure? For someone who doesn't care, they are sure suing Samsung a lot.
Hey Apple, I certainly won't buy an iPad just because you managed to put Samsung's Galaxy Tabs on hold in Europe.
1 1
@albert_swart I'm not even sure there are that many VCs in Berlin. So far it's three universities and a number of startups - and low rents.
Two theories: (1) It's not broken enough to care. (2) We sort of bring this upon ourselves, therefore lack of perceived injustice.
1
Been thinking about the whole peer review discussion. It has been around as long as I can remember, but nothing seems to change. Why?
1
@albert_swart But it's not that easy. What about the big tech companies providing the talent and the VCs for the money?
I'd be as happy as anyone if Berlin became the next startup hot spot. But somehow the recent hype feels somewhat forced. #sorry
Replying to @DRMacIver
Just remember: no alcohol ain't no solution either (as we say in Germany).
(just trying to come to terms with one of the worst mosquito infestations I've experienced in years)
Life's little worries are sometimes like mosquito bites: they itch like hell but scratching doesn't help at all. Eventually it will stop.
1
Mobile web site rules: no multiple redirects (hello, Google), no redirect after loading the page first (hello, Wikipedia). Thank you.
1
Replying to @agnoster
BSD if you don't want to restrict use, GPL if you want to make sure people republish any improvements to your code.
Replying to @antirez
I agree. At the end of the day, you have to actuall do something to get something done. It's that simple... ;)
RT @mariofusco: I am tired of this sterile languages war: good and clever algorithms are far more important than the language in which t ...
Liebes @1und1, das ist nicht euer Ernst, oder? Nach drei Wochen krieg ich das Ersatzgerät und jetzt gibt's 'ne Großraumstörung? #fail
Why is it that I always get the Scala compile server in jvisualvm instead of the real program...
Just realizing that Twitter's notification emails together with Gmail's search capabilities make an OK archive of Twitter conversations ;)
1
Replying to @zoltanvarju
They are huge oversimplifications, of course, but work well if you want to predict something like the topic of a text.
Replying to @zoltanvarju
Very interesting. In my home field (machine learning) people have completely turned to probabilistic models.
@zoltanvarju ... somewhat restrictive to me, given the fuzzy nature of language. But you probably need rules to have exceptions ;)
@zoltanvarju I'm beginning to understand what you meant with generative approach. Breaking everything down to exact rule systems seems ...
Debugging a very odd error in our trending database which only seems to occur on the specific 573214th tweet in our test data.
RT @headius: Common theme talking to Oracle folks at #jvmls: Oracle really cares about multi-language support for Java/JDK/JVM.
Replying to @kreichgauer
So things like elementwise operations and matrix-vector are implemented in Java (because it's faster). Sorry about that... .
Replying to @kreichgauer
Ah, sorry. Because JNI copies everything for each call, you cannot gain on ops which are O(n) or O(n^2)...
Alright. Foursquare had to go to make room for the Google+ app. Sorry, buddy, but LBS are so 2010... ;)
Heyho, kaum drei Wochen später hat @1und1 mir endlich den Austauschrouter zugeschickt. Hoffe, das behebt das Problem...
But I think, ultimately, the stuff you build using the tools is the real thing. #justthinking140charsatatime
Which just makes sense, a (say) messaging framework is different from your main product. Doesn't hurt your business to contribute some tools
One downside of the "companies releasing stuff as open source" model is that we have a lot of tools, but little real applications to hack.
I wonder if the recent rise in spam basically means that Twitter has become mainstream and marketing is now seriously targetting it.
@holadiho Hatte aber ein paar gute Bilder wie "das ist als wären vier Schlüsseltechnologien gleichzeitig erfunden worden"...
I'm not that picky when it comes to followers, but the last 30 or so have only been spammers...
Replying to @smolix
Your account has probably been hacked. Unless you really want to promote working from home...
Replying to @heiko
War da mal auf einem Zivilehrgang. Außer Marktplatz und Donauauen gibt's da nicht viel, glaube ich.
Switching to buffered writes brought the time down from 6 minutes to about one. Hot spots look much saner now as well. #weekendhereicome
I have to admit that I'm a bit excited about the women's soccer final on Sunday. #nadeshiko
Playing around with java.nio to implement my own version of a buffered RandomAccessFile. #someonemusthavedonethisbefore
1
Sometimes, when I'm "in the zone", I think "somebody should really be paying me for what I'm doing here". Oh wait... .
Alright, the new Twitter client or the Goole+ client, what's it going to be? #firstworldproblems
The Facbook/Twitter/Tumblr split was ok, but with Google+ now in the mix, I'm feeling rather confused. Where to post? #firstworldproblems
@ryotat Thinking about tokenization for Japanese search. One question: do you normally search for kanji only or also for inflected verbs?
Replying to @kreichgauer
Sorry, no. AFAIK, number of threads is configured at compile time in ATLAS, so you have to do it yourself. Then use --ptatlas
Ok, that took like 3 hours to set up the tests (and fix a bug the test revealed) and about 1 hour to put in the new feature. #hm
Replying to @DRMacIver
I also considered being able to code this by rote the hallmark of CS education. Then C++ came with STL and destroyed everything.
Replying to @DRMacIver
:) Back then, I used to pride myself on knowing more than one way to deal with the ends (null pointers vs. circular). #iwasyoung
LinkedHashSet, Y U NO ITERATE BACKWARDS? Good thing I wrote double-linked lists every other week back when I was still coding in C.
I do like scripting languages, but I've recently met a few students who have a hard time coding in a statically typed languages.
I'm pretty sure there is a phyiological limit on the number of social network sites you can manage simultaneously. And it's probably < 4.
3
@DRMacIver Oh my, already confused. I guess there is a physiological limit on the number of social network sites you can manage at once.
Replying to @DRMacIver
Apparently, "encircle" is the verb for "friending" on Google+. I hope.
Dang, my HTC desire has too little men (of any kind) to take the Google+ app as well. Sufficient reason to buy a new phone? ;)
Seriously, no real-time search of public posts on Google+? Talk about lack of discoverability...
Also, the amount of kludge your brain can come up if you don't seems to be effectively unbounded.
It's amazing, but nothing helps in getting unstuck as much as drawing a nice graph with all constraints and relations on a piece of paper.
Replying to @zoltanvarju
I will, thanks! Liked what I read so far. Very terse and concise - in a good way.
Replying to @zoltanvarju
Just came across this book in my father-in-laws bookshelf. So I thought "why not, you have to start somewhere" ;)
I wonder whether they will keep their API open or whether they'll eventually shut down everything when they have covered enough services.
Ah, finally found the IntelliJ IDEA customization feature to delay auto-completion. No more "_" expanding to "_RMIConnection_Stub" :D
Replying to @ahans_de
Yeah, I'm already waiting for the embarrasement ;) Someone else sent me an invite, but clicking on the Join button leads nowhere.
Replying to @ahans_de
It seems they're really not adding new people, even if they are invited. Or they're specifically blocking me ;)
Replying to @ahans_de
Thanks, but several people have already tried to add me/whatever, but I always get to redirected to the demo page...
IMHO, Google's real-time search also was a pretty half-hearted attempt so far. Just having big index on the stream cannot be the answer.
"information on Twitter that’s publicly available to our crawlers will still be searchable and discoverable on Google." Crawling != Realtime
Static typing adds complexity because you then need to decide whether to model your relations at compile time or at runtime.
I mean, whole generations have tried Google+, praised it, lived with it, outgrown it, AND I DON'T EVEN HAVE ACCESS TO THE DAMN THING.
1
IntelliJ IDEA is behaving quite erratically. Some days everything's fine, some days it hangs every twenty minutes. Odd.
All those posts and comments about Google+ feel rather surreal given that I don't have access to it yet. #whatarethosepeopletalkingabout
Vielleicht sollten wir die ganze Steuersenkungsdebatte auch einfach überspringen. #nichtschonwieder
My 4yo daughter has a hard time understanding that you cannot just click on the icons on my laptop's screen. Ah, the touch revolution...
1
The plus side of somehow not having access to Google+ is that I'm not distracted from what I should be doing. Well not more than usual ;)
1
RT @flo: RT @weightshift: I have a Life+ invite. All you have to do is walk outside your door and talk to someone. Or that person next t ...
Lately rediscovering Stevie Wonder's classic phase: Talking Book, Innervisions, Fullfillingness, and Songs in the Key. Pretty Awesome stuff.
RT @karpathy: Quadruple checking all my code. I'll just leave this running overnight.... Ah who am I kidding. I *KNOW* I'll come in tomo ...
Replying to @mdreid
... and hoping that there is a shortcut to reaching that level of experience.
Replying to @mariofusco
I guess if you want to comment out larger blocks of code. Of course, why would you want to do.that? ;)
I like the idea behind the Optimistic Weather app. But having to install Adobe Air? 5mb of crashing madness.
Don't you love it when you're debugging and fixing bug after bug just to realize that those weren't the reason your tests were failing.
2
@holadiho Bin eigentlich ganz zufrieden mit twicca. Hat auch was, dass Support wg. der Sprachbarriere grundsätzlich unmöglich ist.
Replying to @1und1
Dank für das Angebot. Erstmal geht's jetzt wieder. Ich drück mal die Daumen...
Replying to @DRMacIver
Oh yes. That explains why otherwise smart people seem to completely loose their ability to judge when it comes to certain topics.
Replying to @DRMacIver
I think in many cases there are compelling arguments for each side depending on context, but the cost for doing both is too large
Replying to @atpassos_ml
IMHO top advantages of static typing: speed, documentation and tool support. Programs can easily fail at runtime.
More silly holy wars: Frequentists vs. Bayesians, Functional vs. Object-Oriented, and vi vs. emacs.
Replying to @DRMacIver
:) Of course, the main reason is that I know the right answer and can't be bothered anymore ;)
In Scala examples, people should rather focus on simplicity and not on what crazy stuff you can do with the type system.
4
I should probably make a list of (tech) holy wars I don't want to participate in anymore. First entry: dynamic vs. static typing.
Alle paar Wochen ist für ein paar Stunden oder Tage die DSL Verbindung extrem instabil. Frage mich, was da bei @1und1 los ist...
After telling one of my (female) coworkers about Justin Bieber she said "I hate him for looking more feminine than me."
Replying to @roidrage
oh yes. I've never figured out how to uninstall with easy_install, for example.
The new Pentax Q looks nice and all, but 1:1.9 just doesn't give you the same depth of field at that sensor size.
Drawing some Scala class diagrams to pull the full extent of my creation into the daylight. #omgwhathaveidone
RT @nealrichter: Wow, either the LinkedIn suggested connections algorithms are hyper-serendipitous or they have interesting third-party ...
Aha, there is a new update for the Scala plugin... Let's see if this fixes these ridiculous bugs.
Also, on every restart of IntelliJ, gnome unity window decorations are gone... . => "unity --reset"
Grrr. IntelliJ's quiet buggy today. Random cursor movements, IDE completely stuck while editing... . Incredible given how old it is!
I have a theory that a programming language has matured only after project names stop having the language specific prefix or suffix.
Zoom's new multitrack recoder R8 looks very nice. Still waiting for thorough reviews to pop up. So far only copy pasta... .
1
Replying to @ChrisDiehl
That was outside some kindergarten where I went for a birthday party with my kids. I guess they had some incidents already.
Replying to @InkmiHq
@codemonkeyism One side effect is that it's easier to work with Java libs than Scala libs. That's why I prefer JUnit over specs.
Replying to @InkmiHq
@codemonkeyism +1 They also changed the serialverid for things like Vector. They're probably just not hardcoded.
Lenovo's online store is really the worst I know. Half of the products are unavailable, and many links lead nowhere.
Replying to @agnoster
Awesome. I might have a meeting at 11, go straight to Leo if I'm not there.
Also, found out that being at work can be quite boring if you're incapable of doing anything useful.
Dragged myself to work yesterday although I had medium severe sinusitis. Much better today. Life can be so good.
Replying to @agnoster
Ich auch. Ich hoffe, wir kriegen dann das schlimmste aus dem Weg geräumt. Ich sag' bescheid, wenn wir die 2.9.0 changes pushen.
Replying to @agnoster
Sag mal, haste nicht auch einen "richtigen" Chat account, das wäre wahrscheinlich einfacher ;)
Replying to @agnoster
It's probably also the problem that you're entering into a rather mature project. Starting from scratch isn't that hard.
Replying to @agnoster
Anyway, just see how far you can get and we should probably continue the discussion tomorrow.
Replying to @agnoster
Every language comes with its set of tools, frameworks, libraries, etc. If you refuse to use them, things get hard, naturally.
Replying to @agnoster
The way I see it, static typing covers only a very small fraction of errors. Otherwise, we wouldn't need runtime errors.
Replying to @agnoster
I really don't know what the all the fuss is about. Honestly, it doesn't take so much to get a clean build env.
Replying to @agnoster
Also, you're wrongly assuming that static typing and test suites have equivalent w.r.t. the stuff they check. They are not.
Replying to @agnoster
The truth is, nothing is ever simple, or clean, or elegant once you go in deep enough.
Replying to @agnoster
Wouldn't be that bad if you finally managed to use the compile server in the background.
RT @ogrisel: The next scikit-learn release will include an efficient svmlight sparse dataset loader and a generic access for the mldata. ...
Replying to @ChrisDiehl
Debugging a program politician-style: Someone always yells "This is complete non-sense, I would never do that". Very helpful.
1
Replying to @atpassos_ml
We had enough serious examples lately: Fukushima, EHEC outbreak, and many politicians are just trying to exploit the situation.
Replying to @atpassos_ml
I see. Another case of people gaming the cost function. Still irritates me a lot.
1
Lately, it seems politics is about making others believe that you're the one to solve our world's problems, but not about actually doing it.
3 1
Wrote little script which plays system sound with aplay when maven is finished compiling. => Immediate productivity increase.
Replying to @headius
Yeah. But they should really take it more seriously. Else people just think "Ah, it's Java, it's slow, it's for the server!"
I really wish the Scala guys would work on their startup time as aggressively as the JRuby folks did.
1
Replying to @roidrage
Also: Writes => "No locks, no seeks". Für einen Read muss man aus allen möglichen Memtables und so die Daten zusammengrabbeln... .
Replying to @roidrage
Das mit den Reads ist auch so ein bißchen Propaganda. Die betonen immer, dass Writes "besonders schnell" sind. Reads halt nicht.
Replying to @roidrage
Die Datenbank war da noch ziemlich leer war. Mit viel Daten und laufenden Compactions sieht das schon anders aus...
Es ist mal wieder so weit: ständige Verbindungsabbrüche bei @1und1. Bestimmt das 3. Mal in den letzten 6 Monaten... #fail
Seeing that half your timeline is at #bbuzz is one thing. Actually being there and wondering what they look like IRL is something different.
RT @manumarchal: Packed room for first European #cassandra hackathon and workshop. Thanks to @acunu @DataStax @SoundCloud and #bbuzz. h ...
Replying to @chl
Yes, I also didn't go. In retrospect, might have been quite interesting...
Apparently half my timeline attended #bbuzz. I'll be joining in on the Cassandra Hackathon at Soundcloud tomorrow.
But how should time management tools have a better idea of what you should do when they are blind to all new developments?
You shouldn't need to keep time management tools up to date. They should actually have a better idea of what you should do than yourself.
Apparently, there is a rather tight temperature range which I find comfortable. 29°C in the office is too hot!
Replying to @agnoster
BTW, you should definitely bring me up to date to what you've been doing lately ;)
RT @gavinpurcell: Looks like Twitter has integrated a Stumble-esque button. Go to your "Following" page and check the upper right corner ...
Really glad that this years #NIPS2012 deadline was on a holiday in Germany this year. Best reason to submit early.
Somehow, Google latitude has stopped updating my location in the background after the last Google maps update...
Alright, one final pass over our #NIPS paper and we'll be done. Paper ID already in late 200s.
Looking at my old code for twimpact I wonder how I ever knew my way around Rails. Too much include magic, no hint where classes are defined.
Replying to @lawrennd
I see. Anyway, I'll have to wait until the price comes down a bit. Let's see how long I can resist the urge to buy it. ;)
Replying to @lawrennd
Oh yes, that would be great. What do you think about its form factor and the stylus?
RT @TorstenBeeck: E-Mail-Tipp: @Boris empfiehlt, möglichst das BCC-Feld als "Abschiedsnachricht" bei Replies zu benutzen. Gute Idee! htt ...
I wonder if there is something like "multi-project bankruptcy" similar to "email bankruptcy".
Of course, being able to discuss learning methods in an application independent way is the reason machine learning exists as a field... .
Discussing applications in machine learning often feels like Zen koans: stripped of all the relevant details.
1 1
Replying to @sscdotopen
The problem are mostly other libraries bringing their own logging dependencies... .
RT @twimpact: Photo: Retweet volume of German politicians on Twitter by party. (Click to Enlarge) It’s not that surprising... http://tum ...
It seem that the hardest part about logging frameworks is where to put the configuration files.
Gehe gerade mal eine Liste von twitternden Politikern durch. Was man da so alles zu sehen bekommt... .
... somewhere deep inside, Java calls hasNext on a Scala List, not knowing that it needs to iterate through the whole list every time...
Cleaning up some 2500+ emails from my inbox, I'm coming to the realization that this whole archiving of emails is pure evil.
IntelliJ is a bit trigger happy with the autocompletion. val => com.sun.xml.internal.ws.developer.MemberSubmissionAddressing.Validation
Scala parallel collections tip: ".seq" gets you back to a non-parallel collection. In case you want to do foreach with predictable order.
1
Putting some serious heat on our clusters with Scala 2.9.0's new parallel collections. List(1,2,3).par -> parallel map/filter/etc.
1
"Single Linkage Clustering". Haven't read this term in nearly ten years. Always good to go back to one's roots ;)
1
It's so weird actually calling someone with my Android smartphone. Just thinking "why am I holding this mini-tablet's screen to my ear??"
1 2
So... hard... to... focus... on... just... one... thing. One day, this sustained multi-project multi-tasking will make my brain split.
Every time I think I'm through with changing the world, @lemire posts something insightful to remind me we should at least try.
1
RT @nealrichter: When #bigdata means simple lots of repetitive data, one should think twice about boiling it all in hadoop just because.
Replying to @mdreid
At least you're trying. ;) For some conference deadlines are the ONLY means of time management.
@ideanado I also couldn't find out. If you look around the web, most people think that 85°C is already pretty high... .
Hm. I should probably try to find out what the maximum temperature for a CPU is. 85°C looks pretty hot... .
Doing some thinking while ATLAS is compiling. Wondering why C compilation traditionally fills up 1000s of console pages and Java doesn't.
Replying to @DRMacIver
Grumpy at least with geolocation. This must be the new millennium ;)
Talk went fine. In fact, just finished on time down to the minute. Years of teaching and giving talks finally pays of ;)
Bradley Kuhn said that MySQL sales people used to tell companies that SQL stored procs violate the GPL to sell licenses. That's pretty sick.
Writing some last minute Ruby scripts to extract performance figures from messy logs for tomorrows Cassandra talk.
Coder's occupational disease #3: Believes that every communciation problem has a technical solution. Team doesn't work => set up a wiki!
Finally learned how to remove subprojects in IDEA: Do it in the maven pane, not the project structure. (Thanks, @thinkberg)
RT @ogrisel: New in scikit-learn 0.8: Hierarchical clustering, kPCA, Non-negative matrix factorization, partial Python 3 support http:// ...
Replying to @ChrisDiehl
I'm sure you will! I've been thinking a lot about this as well. We'll see... ;)
Programmer's occupational disease #4: Failing to understand that in the real world, rules and protocols are subject to interpretation.
Replying to @headius
In the mean while I learned that Rhino also compiles to bytecode. But probably it lacks the right kinds of optimization... .
Replying to @WorkFlowy
Sure, why not. Not sure what the state of my Skype installation is, though ;)
Replying to @headius
Great idea! Bring JavaScript back to Java! I guess Rhino only interprets, doesn't compile to bytecode, right?
Replying to @ChrisDiehl
Congrats! That reminds me of the conversation we had earlier this year. So you made your decision to change environments?
RT @auerbach: Ads on news sites don’t get clicks: “...in no case did five people click on the same ad...in the months studied.” http://o ...
I really like WorkFlowy's ease of use. Now if only someone could add a timeline, velocity and burn-down chart ;)
Oh my, someone already put up a 3d model of Osama Bin Laden's compound on Google Earth... .
1
Unless the window is already at the right edge, then the controls just appear under your pointer. As it should be, really.
The new Gnome scroll bars are quite awful. You first have to hover to make the controls appear, then move to the right to use them. #unity
To all you overcommitted guys: Remember it matters what you complete, not at how much you fail. At least that's my hope... ;)
1
Lucene is really the apex of Java's "some assembly required" approach to designing library APIs... .
@ideanado_ Actually, I think what annoys me most is that people assume that just because I use Twitter, I'll defend it, too.
@ideanado_ That's probably my problem. It's really hard to explain to someone who hasn't experienced it himself.
Replying to @sscdotopen
Cool! If you want, come visit us some time and tell us what you're working on!
Two discussions I don't want to have anymore: (1) Do you really need a smartphone? (2) I never understood what's so great about Twitter.
Ok, after YEARS of focus-follows-mouse, do I now have to change to click-to-focus just because Unity moved the menu bar to the top?
Doing some strictly in-browser organization stuff while my laptop is upgrading to Ubuntu 11.04. Unity, here I come!
RT @twimpact: updated beta.twimpact.com to include a tweet cache and see more pictures with 24 hour activity #twimpact #trends
Had to find out first whether Delicious opt-in email was legit. It is! Delicious becomes next YouTube! (sort of)
I'm very much in favor for very simply structured review forms. Paper summary, positive/negative points, recommendation. That should be it!
1
Replying to @munmun10
Have you tried paper.li? Not exactly what you want, but updates daily.
Slowly recovering from an illness I got from my 1yo son (thanks, pal), spending most of my time on tvtropes and memebase. Ah, the internets!
1
Either I'm onto something or programming in dynamic languages has permanently ruined my expectations to static typing ;)
The last few days I ran into some issues with generic types in Scala where things didn't work out as I expected. I'm mildly worried.
Really interesting to read Evan Williams' story on Blogger in "Founders At Work". Apparently, the interview was before Twitter.
I had a dream that someone hacked my Android phone and every time I tried to unlock, I had to play some puzzle mini game. What a nightmare!
Wonder why IDEA is a bit more flaky than usual today... . Refuses to take any more keyboard input at some point.
Replying to @nikete
Thanks for the link. The problem is that I need this to work with our Scala codebase, so no short-cuts through existing libs... .
Still working on language detection. Realized that you need a whole lot of infrastructure to use an SVM, e.g. for tuning and validation.
Waiting for our Shogun GSoC phone conference to start... Experiencing technical difficulties. ;)
Replying to @ogrisel
Interesting, I see you're using tf/idf and an SVM. Does the tf/idf improve a lot? I'd expect the SVM to take care of the weighting.
Apparently, training a language detector with Huckleberry Finn gives suboptimal results on tweets ;) No "Video", "Robert Pattison", "Bieber"
Replying to @sscdotopen
Nothing fancy, just n-grams (on the character level). Did some normalization by converting likelihoods to quantiles, though.
Replying to @mariofusco
Also, I think much of the more complex stuff is not really necessary to get started.
Replying to @mariofusco
You're right! IMHO C++ is much more complex and confusing, for example.
Computing confusion matrices for twimpact's new language detector. Danish is getting a lot of false positives. I wonder why... .
1
First day back in the office after our @PASCALNetwork Steering Committee Meeting at @CumberlandLodge. Nice stay, excellent food, as always.
RT @PASCALNetwork: CfP: Google Summer of Code - Shogun Machine Learning Toolbox (Deadline April 8, 19hrs UTC): Call for Participati... h ...
Thinking about it, that's probably true for all languages with some form of type inference.
The longer I work with Scala, the less I see the case for dynamically typed languages. They really made static typing much less painful.
Prototyping some simple n-gram model based language detector for @twimpact. Thanks to @gutenberg_org for the texts used for training.
Replying to @ChrisDiehl
I think it's pretty remarkable that one of the most important data sets in was a side product of litigation.
Make your company's name immortal by donating a data set to social network research (And try not to get sued over it like Netflix did ;) )
1
I'm always amazed at how dominant the Enron data set is in social network research. I fear it's the lack of similar realistic data sets.
.@thinkberg has me worried. Just wanted to show me the really cool JavaScript code he wrote.
RT @jonasabreu: @mikiobraun it uses some backdoors, like instantiating an object without constructor invocation. You can find this at ob ...
Still, how can you construct an object using a superclass's constructor? Some magic happening in sun.reflect.*, apparently.
Replying to @chl
"if the class is serializable, invokes the no-arg constructor of the first non-serializable superclass." Not sure what that means... .
Replying to @chl
Yes, that's what the documentation says. But actually it works even if there is no such constructor. See ObjectStreamClass#newInstance.
Letztens sagte mir einer von @1und1, die hätten doch die beste Technik. Wieso klappt's dann gerade mal wieder überhaupt nicht?
Replying to @sscdotopen
yes, but that only works if you have a constructor taking no arguments.
Already learned how to read private fields. But how do you create objects without using a constructor?
Still obsessed with Java serialization. Wonder whether it really is implemented using only reflection and no backdoors.
Replying to @chl
Wie wäre es mit einem sechsmonatigem Moratorium der Sommerzeitregelung? ;)
Hm. Java serialization doesn't really fit Scala which has a much clearer idea of object creation and what the mutable state is.
Adding serialization to one class spreads the need to do the same like a virus. But I guess we all know that... .
Replying to @peteskomoroch
@bradfordcross I bet I can produce similar plots from random data...
Replying to @peteskomoroch
@bradfordcross That cited paper didn't impress me at all. Look at Fig. 3 which should show segments with correlation and w/o.
Replying to @peteskomoroch
Thanks for sharing that article (although they consistently misspelled your name :( )
Quora feels a bit like Jeopardy: you talk about any topic as long as it is formulated as a question.
1
Replying to @davidandrzej
I think it's about understanding which part is essential and why.
Typical failure mode observed reviewing: Methods with lots of deliberate design choices seems to perform well, but little insights into why.
I'd really appreciate if the Microsoft CMT supported theming. I'm getting confused because all conference's review sites look so similar.
Replying to @chl
@pbrane Although eventually you would want to integrate things like individualized trending into "core clients".
Replying to @chl
. @chl To clarify: you can still work on those things, but you cannot integrate them in a 3rd party client like TweetDeck.
Oh great, one of my two 4GB modules is faulty on my Lenovo laptop. Down to 4GB now. Let's just hope the file system hasn't been compromised.
Runtime extension for German nuclear plants has been temporarily suspended. I'm not even sure I could write that down in formal logic.
1
After reading latest Twitter API Rule changes decided to stick with a non-official Twitter client. Just so. #twicca
2
Luckily, all of my Japanese relatives are ok. Still, knowing that a nuclear reactor is just blowing up a few hundred km away is just awful.
Weather forecast for today: 12 C. Will this winter finally come to an end? Got really cold already in November.
Twitter web page should also integrate @-mentions with retweets like the Android app. Easier to see what people are reacting to.
Unfortunatly, the ATLAS which ships with jblas 1.2.1 seems slower than the old ones. Recompiling ATLAS on everything I got to check.
Contemplating what Android's "mem full" notification icon - constant companion by now - is supposed to show. Hard disk with aux wheel?
RT @Nico: Verweigere GEMA und Plattenfirmen den Zugriff auf deine Website oder deinen Blog. Sie blocken uns. Jetzt blocken wir sie http: ...
Replying to @ChrisDiehl
Stream on the left is pretty much as tweets come in. We're using Twitter's track streams, so we're seeing a random subsample.
@thinkberg Das gibt's doch gar nicht! Muss er jetzt seine Bibliotheksausweise zurückgeben?
What Twitter is missing is the equivalent of a party: a time-boxed event where you can talk with your friends' friends.
1
RT @roidrage: MapReducing the shit out of thousands of Justin Bieber tweets, because why the hell not?
Anyone looked at StreetView in Google Earth recently? Seems they added some basic geom. reconst., mostly leading to psychedelic effects.
Realized just yesterday that IJCAI reviews were also due this weekend. Down to two reviews left now...
Reviewing papers for ICML. Reminds me that although we often complain about peer reviewing, reviewing papers can also teach you new stuff.
1 2
Muss mal gesagt werden: Abschnittsweise Text behutsam einarbeiten != Fußnoten/Anführungszeichen vergessen. #guttenberg
RT @CumberlandLodge: @mikiobraun Some of #KingsSpeech was filmed here-well spotted! Next time @PASCALNetwork are here we'll give you a t ...
Saw The King's Speech yesterday. Could swear some of the shots were taken at Cumberland Lodge where we have our @PASCALNetwork meetings.
1
Replying to @roidrage
I found it quite interesting just how different those programmer personalities were.
Replying to @karpathy
;) On the plus side, I was actually quite impressed that they pulled of WinPhone 7. Nokia had been struggling with that for years.
Replying to @InkmiHq
@codemonkeyism Oh yes. I think one cannot have escape characters in triple-" strings. Not sure, though...
Replying to @karpathy
Oh yes. Individual installation/update/removal procedures for each and every software package feels so archaic. It's 2011!!!
Not to mention the 54 updates Windows is installing now... . Hey Microsoft, how about some central repo of software like... an app store???
Heck, every time I boot back into Windows after a few weeks of Linux only, I have to click through dozen of update requests to get working.
Replying to @naomimeran
Ich sag immer: es ist ein Wunder, dass Facebook überhaupt funktioniert.
Methinks I cannot resist those new Android Honeycomb tablets forever... . Luckily, none are on sale yet.
Oracle vs. Hudson etc. shows pretty well that Open Source != openly available source code. It's about the process and the community.
1
Weiß gar nicht, warum Elternzeit abschätzig Wickelvolontariat genannt wird. Kinderkrankenpflegerpraktikum trifft's viel eher. #kindkrank
Nothing compared to what is really going on in Egypt right now, of course. Hope everything will turn out fine.
Replying to @karpathy
I can wholeheartedly agree. This is actually the first game I ever played through twice. ;)
1
Soso... @1und1 hat angeblich "noch nicht offizielle" Großraumstörung in Berlin. Hab ja zum Glück mobilen Hotspot dank Froyo ;)
Doing some pair programming with @thinkberg for some new nifty tracking feature on @twimpact.
I do like the idea of immutable data structures. But why is iterating through a Scala TreeSet so slow???
RT @ogrisel: I like my confusion matrices with a deep blue background and a bright red diagonal in the middle.
Playing around with GotoBLAS2. Very fast. So many architectures to choose from, though. Have to rethink how to package for jblas.
1
Just signed my contract addendum for working part-time the next two month to spend a little time with my 1yr old son. #thrilled
What happened to MongoDB? RT @joindiaspora: We're migrating our database over to mySQL. Should be back in under an hour.
Finished reading the Cluetrain Manifesto. Hard to believe this predates social media. Otherwise we're still at business-as-usual, I'd say.
Always a good feeling to leave home with a fully charged smartphone, confident that you won't have to recharge it before evening. #its2011
RT @munmun10: My apartment complex not only has a Facebook page, but somehow they ended up sending me a friend request. This is totally ...
Low priority task in my brain notified me of a new idea to speed up trending for @twimpact. Will have to wait till Monday to check...
1
Aha... Taylor expansion is your friend. I guess that is the first time Taylor expansion helped me to understand the real world.
Hm. I've got this really complicated formula here and when I plot it, it's a straight line. Where is that table with exp and log equations?
1
Extremely high density of Powerpoint slides in todays Ph.D. seminar. With math. #whatswrongwiththosepeople
RT @ChrisDiehl: Heavy dose of amazium needed RT @dangerroom: Darpa Wants a New Mathematical Language to Explain... Well, Everything http ...
As usual, as soon as thinking about going home, code starts to flow freely from my fingers... .
Hm. Looks like the author of "manage your project portfolio" expects you to have read her other book "manage it!"...
3 operating systems, 4 programming languages, 2 C calling conventions, 32 bit & 64 bit support. That's what I love about jblas ;)
Alright, doing some tests in C on a text console with Makefile & emacs just like it's 1995 to find that jblas bug... .
Oh great, after migrating jblas to maven and finally running tests in each build, discovered errors in basic BLAS routines on Macs...
RT @ogrisel: My #PyCon talk "Statistical machine learning for text classification with scikit-learn" got accepted! abstract here: http:/ ...
Trending is the new Search. With real-time, search is just too static. Do you know any site which has informative trending topcis?
Replying to @ChrisDiehl
I think that pretty much describes the situation I'm finding myself in right now. :)
Replying to @sbtourist
Given the licensing issues around BDB for commercial use, I would in fact welcome that.
Replying to @ChrisDiehl
That's too kind. In reality, it feels more like a constant conflict between what I would like to do and what I should do.
Replying to @ChrisDiehl
Of course, as always, it's unclear how to improve this situation. In the end, only people setting the policies have the power.
Replying to @ChrisDiehl
I think many are even unaware of how much their focus on getting published interferes with their choice of topics to work on.
To create a better trending algorithm for @twimpact such that I never have to see Bieber tweets again. #newyearresolutions