fnl

36 ms·

331.
by fnl 11y ago
Skip-grams only take focus word, context word pairs. So those pairs do not take order into account
332.
by fnl 11y ago
Quite exciting, although the tradeoff with respect to the number of rows against the speed-up could be relevant. I.e., in the micro-benchmark being shown, how does the choice of table size affect the outcome of the benchmark?
333.
by fnl 12y ago
Too bad this was missed...
334.
by fnl 12y ago
Tyrol never was German (except when anexed by the Nazis, that is...). At most, it was part of the Ostrogothic kingdom (up to 550 or so); After, it was its own county, ruled by counts apointed by the Holy Roman Emperor, and then (1369, with
335.
by fnl 12y ago
By the way, regarding your other points (although I think they are a bit off topic): As for the cost of corpora: That number is extremely variable, and in many cases (outside of newswire, in particular) you might have your own training data
336.
by fnl 12y ago
I ran the systems over the CoNLL 2000 PoS data (see training time), consisting of roughly 10k sentences, split into 8k training and 2k testing sentences (by the organizers of the ST). However, this review is not about "who has the best
337.
A review of free sparse sequence taggers for NLP(fnl.es)
8 points by fnl 12y ago|3 comments
338.
by fnl 12y ago
Well, teddyh might have a point here, nonetheless: By now, I understand that ftfy is about fixing mixed up encodings between UTF8, latin-1, CP437, CP125[12] and MacRoman (only). But by claiming you are fixing "Unicode" in general
339.
by fnl 12y ago
Too bad this was missed by the HN crowd. IMO, a nicer monad implementation than fn.py.
340.
by fnl 12y ago
ever since we have an induction stove, we're technically "e-kettle free". so I could state the contrary: "what? you still own and use an e-kettle?" ...
341.
The worst registration "CAPTCHA" question ever (SciDB)(scidb.org)
1 points by fnl 13y ago|1 comments
342.
by fnl 13y ago
And so yet another reason for "not migrating" to 3.x is off the table...
343.
by fnl 13y ago
That's because, I may presume, you have never (really) used make. For somebody who does, exactly the opposite will be true.
344.
by fnl 13y ago
Exactly my thoughts too, after reading the ageism article. An "old guy" (like me) would not even think twice and just "make" great use, as he has always been and with great success nor the time-consuming need of learning
345.
by fnl 13y ago
or in the want of more interesting news...
346.
by fnl 13y ago
So the size of the dots are the Money_JPY "volume" - that is, Yen, and not Bitcoins? Isn't that a bit redundant, because it contains information from the y-axis, which encodes the Yen value of a BC? Wouldn't it have been
347.
by fnl 13y ago
Turns out, atea does all I need and integrates nicely into the OSX UI with an icon in the menu bar. It does: - project-based task management - proiority ordering of tasks - Dropbox-based syncing of one or more task files - time tr
348.
by fnl 13y ago
Seems I have found it, at least for OSX: https://github.com/pkamenarsky/atea Not perfect, but as close as possible
349.
by fnl 13y ago
Nice; I'd like something like that, but with the addition of allowing me to track the estimated vs. the real time of a job. I.e., when I start a task, I estimate the time needed, when I finish it, it tells me the real time it took. Ide
350.
by fnl 13y ago
I am happy to have seen Ioannidis' talk on this very matter a while ago - very inspiring indeed. However, I think, often such articles as the one in The Economist lead "outsiders" to the opinion that much of the current biome
351.
by fnl 13y ago
I could not agree more to your description of what the life after your PhD seems to be. The only other workday activities you are missing out are probably (a) travelling, meetings, and conferences [lots] and (b) reading papers [hundreds&#x2
352.
by fnl 13y ago
> the "insight into the world" comes from physic, astrology, etc. then I hope we live in two different worlds - preferring astrology over religion will not get us very far, I guess... (just cracking some balls, but it's a
353.
by fnl 13y ago
Then I don't get it: According to what we can see, in March 2014 they already should be "hitting it" (1 year after the initial March 2013 spike), but it looks pretty bleak to me.
354.
by fnl 13y ago
Seen quite a few times (NLP web APIs), and my opinion is that this kind of stuff tends to not be scalable: to be useful, such web APIs have to be able to do entire articles in just a split fraction of a second. Although I am not sure (becau
355.
Birdsnap: Guide to and Image Classification of Bird Species(birdsnap.com)
1 points by fnl 13y ago|0 comments
356.
by fnl 13y ago
From the article: > We show that sex differences in gene expression and splicing are widespread in adult human brain, being detectable in all major brain regions and involving 2.5% of all expressed genes. While I don't have a simila
357.
by fnl 13y ago
This is missing the universally true #1 mistake probably (nearly) anyone commits when starting in ML: Missing an excellent understanding of the problem/domain (Unless you happen to be a domain expert for the problem you are working o
358.
Honnef.co/go/implements - identify interfaces implemented by a type(github.com)
1 points by fnl 13y ago|0 comments
359.
by fnl 13y ago
Like the C vs. C++/Java bridge - certainly more succinct answer than my reasoning... :) But I certainly do not believe that Scala.js has a reasonable chance of having an impact alike GWT or such; maybe it can get close to (ruby) Opal o
360.
by fnl 13y ago
I had been asking myself the same question over the last few months, maybe even longer. The final selling point for me was lines of code, caused by the feature-richness of Scala. While implementing a few generic data collections and algorit

More ›