Dez 14 2009, 22h29TODO: faire un top 10 des meilleurs dessins animés crétins
Out 31 2009, 21h45
Abr 2 2009, 13h05Amazon announced today that they have released a tool to make it easy to run mapreduce jobs on EC2:
For EC2 users, this change makes things way easier than before - EC2 had little to offer so far when it came to automatically deploying executables, starting and monitoring jobs on multiple instances, not to mention fault tolerance. A couple of weeks ago I went through the painful exercise of starting an Hadoop cluster on EC2, and it was not exactly what I would call an enjoyable experience. Now you just put your data and binaries/jar on S3, and get rolling. Wowwowow.
I've run mapreduces since day 1 (literally) at Google, and I am delighted to see that the ability to efficiently process massive dataset is now accessible to anybody - being able to do so was - personally speaking - an enlightening experience, and I am glad that this enlightenment is now available to everybody. There's more to it. I personally think this release is going to open the path for a new way of building intelligent web applications.
Take the example of sites like flickr or last.fm, which have repositories of data that do not really belong to anybody. Flickr and last.fm understand that, and they want users/developers to benefit from it - so they let developers access this data through their web services (instead of having the developers scrape their pages like pigs >:-|). Wanna know bob's favorite songs? Get it from the last.fm web services. Wanna get the urls of bob's thailand trip photos? Get it from the flickr API. Now do the slideshow mashup... The trouble is, those APIs are good for mash-up "join"-like applications, but do not really allow for more intelligent apps. What if I want to build a slideshow generator that needs to retrieve flickr photos by color or any other content-based features? What if I want to build a social app on top of last.fm, or a visualization tool, and need to aggregate data from hundreds of profiles? The data is there - but it's not by making hundreds of calls to the web API that I will efficiently get it. Webservices are about repurposing *records*. What if we could also repurpose *datasets*. So what could last.fm and flickr do to help developers build cool stuff beyond mash-ups? Dumping a snapshot of their publicly visible data to S3, and letting us, developers of intelligent indexing/content-repurposing apps, run mapreduces on it.
Now take the example of the Echo Nest. They have this nifty web service in which you submit a song, and get an analysis XML file with plenty of interesting features related to the music content. Mashups built on top of it include song visualizations, song remixers tools, but I haven't seen a single content-based music search service or playlist generator. What if I really want their analysis for my 5000 songs corpus - but I am not comfortable with moving my data to their servers (bandwidth, cost, and maybe copyright reasons)? A web API feels awkward for that. What a developer of such an app would need would be a way to ask the echo nest: offer me a way to run your binaries (and let me use as many machines as I want - after all I am paying the bills!) on my data.
What I have in mind is a market in which owners of massive corpora [research institutions, web sites with huge public databases] give (sell?) access to their data; and in which innovative media analysis or data-mining companies give (sell?) access to their analysis tools. Up to developers to mix and match. Let me run my tool on your data, let me run your tool on my data, let me run X's tool on Y's data and join it with Z's data. This will require trust: the providers of an analysis tool won't be pleased if third parties have access to their executable and reverse-engineer it. If the access to a database is protected, the owner won't want people to run dumb "copy" jobs on it. But if all this occurs through a platform everybody trusts, with the right policies and access restriction mechanism, this could work. S3/EC2/elastic mapreduce, and a missing product doing the plumbing/billing/ACL management could be such a platform.
I don't even mention the possibilities of such a platform for doing scientific research - it makes me cry that I did not do my Ph.D. in such interesting times.
Ago 31 2008, 17h44I had a good laugh reading this:
"the mp3 format mutilates sound quality and your hearing"
To the author mp3 = compression ~ dynamic compression (it is not clear if the author gets the difference, he describes dynamic compression as something reducing bandwidth, and somewhat considers that the two go hand in hand) = loud music = danger for public health. Oh, and mp3 is also held responsible for a) young people not dancing anymore at parties ; b) people buying less and less high-end hi-fi equipment ; c) people listening to music while doing something else (crime!).
Ha ha ha.
Before mp3 and the "loudness war", people were blowing out their ears with their walkman or crappy PA systems in clubs and concerts. Before mp3 and digital audio, FM radio stations used compression to attract listeners, and mastering specialists had to follow to recreate the FM sound on CD. Without mp3, we could still store thousands of tracks on an ipod, and have the same listening/music hoarding behaviors. mp3 quality sucks, but, *breaking news*, most people don't give a shit about the quality.
Let's ban nightclubs, concerts, ad-sponsored FM radio, Nielsen ratings and mastering plug-ins. But let's keep the mp3 format around, because otherwise the only way I would be exposed to new music would be through crappy FM stations or "long tail, what??" record shops.
Mai 2 2007, 14h40
Abr 19 2007, 19h17The first one looks like a nabaztag
Track : Split Screen
Abr 19 2007, 11h01
Mar 25 2007, 16h41
Jan 26 2007, 15h21LOL
Waiting for the results of the audiofanzine drawing contest...
Jan 24 2007, 23h55Description
Bonjour ! Je suis Gédéon, je suis un Cupidon. Epicurien romantique depuis peu, je débute dans la poésie, mais j'ai déjà appris par coeur dans mon coeur 127 extraits de poèmes mis à disposition par la RATP dans les wagons du métro. J'en connais avec des phrases où les mots finissent pareils et ça sonne comme des chansons. Je vous les déclame en chantant que pour vous dans les yeux au restaurant avec le café, c'est trop la classe. Libertin, je m'attache pas et vole dans les airs de fleurs en fleurs -vous Mesdemoiselles- tel un papillon aux couleurs soyeuses avec un léger bruit de bourdon en orbite. Je suis très câlin et n'ai qu'un désir, déposer ma semence par-ci par-là à défaut d'autre chose. Je suis à l'Amour ce qu'un intégriste est au mysticisme ou un communiste à l'espoir social. Carpe Diem, sole meunière ou truite saumonée, je m'accommode à toutes les sauces, même les plus rustiques, car je n'ai qu'une passion : LA FEMME. Aucune croupe n'échappe à mon oeil de photographe. Je suis le roi du cliché, spécialiste de la carte postale Notre-Dame en bateau-mouche la nuit en été. Ah ! le sourire d'une femme ! Son plaisir est le mien, mais j'aime mieux quand c'est le contraire. Elle reste pour moi un Mystère, le plus beau des mystères car de toute façon, je ne comprend rien à rien. Je t'invite donc à voyager avec moi après le boulot le temps d'un instant, deux peut-être pourquoi pas. Suis-moi dans les airs, bondissons ensemble de nuage en nuage. En-dessous il pleut, mais là-haut le ciel est toujours bleu, les rires sont bleus, bleus seront nos câlins. Je serai ton Apollon et toi mon Aphrodite, ton Damien et toi ma Cécilia, viens, viens, oh toi ! Abandonne ton tricycle et viens sur mon tandem. Partons de zéro et ne soyons plus qu'un. Faisons un six et un neuf dans le grand huit du ciel (ami lecteur, tu as là au passage les six numéros plus le complémentaire du prochain loto ainsi qu'un exemple concret d'évolution dans un ascenseur).