haddr

It's not a secret that in the process of automatic manufacturing there are terabytes of data produced in a matter of minutes. OF course how many of such data is of any use is another story, but anyway it still requires storage and processing power capable of handling that volumes.

I think that at some point this O(n * sqrt(n)) is actualy not precise. Maybe it works for the first few GB, but then other mechanisms come into play.

For example processing 100GB of data actually don't have to be O(nsqrt(n)) because if you process it on cluster, then other machines are also using L1, L2, L3 caches and RAM. Then the whole process can be streamlined which means that some operations can be faster than the pessimistic nsqrt(n).

Spark is somehow crucial in traditional car engines...

Please check the special $1000+ power cables for audiophiles...

This! I rememebr back in 00' we were always talking with friends about creating such map (school, neighbourhood, etc.) because it seemed funny to play on such map. Unfortunately nobody was skilled enough to make such map...

This is one of those few moments when saying Big Data is actually a well-used term.

Only applied to image recognition.

Yep, people are really forgetting that most of the time it's just being clever rather than having (or using) in-depth knowledge is what allows to bring some good solution to the table.

An example from today: we have a very slow pattern matching code, that starts to be a bottleneck for the application. What can you consider? Well you can dive into sexy bloom filters, experiment with some Trie-based structures. But then when you analyse the problem it results that simple word lookup with a simple hashtable is the fastest solution for given constraints. No big deal, no rocket science.

Probably the same goes for rocket scientists, but one level higher ;)

It's not key study, as title suggests.

I learned that you can grasp English without learnig grammar (this is what they do in UK). But while it works for some people, it might not work for others. And it definitely wouldn't work for some more grammatically complex languages...

I think 'lingua franca' is never used in a purely literal sense, so probably nobody even thought about what you mention (which is definitely true)

This article is cool, I love the way is actually encourages to learn something by starting with something simple that you can grasp in seconds.

I was like (｡◕口◕｡)!!!

Side note: the hbr.org website is taking over 30% of the screen by showing this thick upper gray bar, which the only function is to show you how far you've already read. What a waste of space & UI.

This thing appears on their social network too: https://www.vica.io/our-story/

What a creepy story

I got very nice error from their search engine, revealing a bit where they get results from :)

http://pastebin.com/RnsB33Ea

Some things should be probably updated on that page, I think the Sonoluminescence problem has been mostly explained.

In Europe Olive oil is also one of the most faked type of food.

Can I upload such image for instance to facebook and intent to run it as html (with some JS inside)?

It's a very nice read with many good points, but any person with some experience in IT projects could argue with it. The author is taking one side without any self-criticism.

It is true that configure scripts are probably doing some useless things, "31,085 lines of configure for libtool still check if <sys/stat.h> and <stdlib.h> exist, even though the Unixen, which lacked them, had neither sufficient memory to execute libtool nor disks big enough for its 16-MB source code", etc. But then what is the alternative? Writing a configure module each time by every programmer who wants to release some software? This is called code reuse and yes, it's not perfect but it saves time. By not reinventing the wheel again and again. By reusing something that is stable and has been there for some time. Probably such thing is generalizing over many architectures, and making useless things, but then again, who cares for some extra 5-10 seconds of the "configure" command, when you are covered for all those strange corner cases that it already handles?

That's actually a middle dot, not a regular dot.

That was the point precisely

I tried it, the GUI is veeery slow for slightly bigger text files. Also it is missing word wrapping. Good luck in diffing some latex files with it...

Ok, checked this one against my diffs, and here's some observations:

+ eskill has a good set of diff algorithms, char level, word level, etc.

- lacks line wrapping

- a bit sluggish (but it's not that bad)

Would recommend trying.

I was recently looking for a fast diff gui, one that would:

- work well with longer files and long lines (i don't know why many of them are very slow on larger files)

- have line wrap option (quite unusual it seems)

- have diff ignoring whitespaces (you would be surprised that not all diff guis support this)

- the best would be to use the same algo that "git diff" does

- if not then any other that works on character-level.

Of course you may use the vanilla git diff, which is really great (even outside of git repositories), but it is not as handy as GUI sometimes.

The closest GUI was meld, but it's not perfect either... macvim with some diff plugin was also good, but not perfect.

What is the meaning of the dot in the big O notation there? like O(m.n)? Should it rather be O(m*n)?

I've sent you an email

sure, how can i send it to you?

For supervised classification this tool is suitable when your dataset is large enough. I performed some tests with binary classification (twitter sentiment) on the corpus with ~7.000 samples and the result is not impressive (~0.77). Vowpal wabbit performes slightly better here, with almost the same training time.

I'm looking forward to try it on some bigger datasets.

I also wonder if is it possible to use separately trained word vector model for the supervised task?

Thank you!

How does it compare to the state of the art methods that used those datasets? Is it better or comparable?

Would anybody explain me what is the takeaway from this approach?

When I run the code I see something like: 38/100 = 42.0% - 38.0% accuracy, 90.47619% accuracy of rated data.

90% is nice, but it looks to me like a tradeoff between precision and recall, where we just label those samples with the highest probability of pos/neg...