data_scientist

Awesome! It looks like we could extract the depth from the video, but my knowledge of ffmpeg is too limited for that. Do you know a way to do it?

Really nice! I would love to add all my notes here, but I sometimes needs to include images, and this is a blocking feature for me. Any plan to include that?

A really nice to have would be to have inline Latex mathematical formula, but it may be too specific for your app.

I needs to connects to multiple servers at the same time, and tmux sessions shouldn't be stacked. How can I do that without tabs?

A short optional sample of every composers and songs would be nice, just to get an idea of the music. I can't remember music in my head (kind of Aphantasia for music), so it's really hard for me to choose just from the list.

What about using machine learning? It sounds like a discovery issue, and recommender systems work very well for that (eg. the side bar of Youtube).

I just discovered Neocities BTW, it sounds very interesting!

It doesn't have a good accuracy. I have yet to see a real-life dataset where it's better than to just call LogisticRegressionCV from scikit-learn. For bigger datasets, you may use vowpal wabbit or Fasttext. It may be a little bit slower for the training (but not so much), and as fast as LR for the training. What is the purpose of using an algorithm when another one is just better?

Does anyone have a good book for advanced linear algebra? I learned the basics in school - eg. Matrix diagonalization, svd) but I need for ml to really understand things like angle between flats, tensor decomposition, etc

SEEKING WORK / REMOTE or Paris France

I'm a data scientist with experience in various types of machine learning projects. I have a statistical background, and strong SWE skills.

I have experience in structured data, NLP, images, deep learning, recommendation systems, predictive maintenance, etc.

Why should you hire me? - I can understand your problem, torture your data and find the right model - or I will try something new, out of the box, if you need it - I know how to work with best practices for the code and how to integrate into a workflow - I love to deliver quality work

Drop me a mail! matthieu@databiz.io

Nice statistical property: this is the same thing.

P(B|A)>P(B) <=> P(A|B)>P(A)

My main issue with R^2 is that it is an artificial indicator. It has no business meaning. The best alternative is to build a metric that is related to your business issue. How many $ do you loss if you're wrong by 1%? Or by 10 units? Do your business loss is linear in your error? Is it symmetrical?

These are the kind of questions you have to ask yourself. Defining a metric is hard, and there is no good shortcut.

Great idea! Some questions:

- Both R and Python support strings, factors, and complex objects in a dataframe. What is NOT supported by feather?

- Feather is "not for long term data storage". Will it be standardize in a distant future?

- Do you plan to integrate it into Pandas?

PyMC 3 implements HMC. It is still in beta but quite stable

Relevant XKCD https://xkcd.com/810/

How do you handle strings on hdf5, on an efficient and portable way? I didnt find any convincing answers.

Genuine question, why do you find Pymc3 better than stan? Pymc3 is still in alpha, while Stan is stable since many years. They both implements the same algorithm (NUTS), so is it just the nicer syntax or is it anything else?

Please modify the bindings, the directional keys are international, without a qwerty keyboard I can't play - and no, I will not modify it systemwide for just a JS game.

Nice one nevertheless!

I just saw a tweet of Notch

> Turns out unrolling tight inner loops really speeds them up! 38% speedup from that alone. Makes sense, it's doing one eight of the branching

https://twitter.com/notch/status/595509436040491008

Even billionaires may want to micro-optimize! Bonus point: he is developing for a "20 years old platform"

I can't find anything about it in English. Do you have any link?

Please don't make it configurable, because it will mean there is no longer "a" YAPF standard. Also look at Pycharm reformat tool, it would be a shame if both couldn't be used in the same versionned project.

One useful feature is to consider the neighbors rows, for example for differentiating a time series. Do you plan to make an efficient method for that?

This problem is addressed by numexpr, which also handle multithreading.

Minecraft online also have all the properties you cite, and there are even more options, collectiond, and human interactions. Think about it: the other players can be adversary or allies, and that can change whenever you want.

This is the bad part of HFT, the one that is theft and destroy value. It is paid by the big players, and "solved" by imperfect solution like dark pools. A better solution could be to add a hour component to the order, so all commands from an actor in all markets are executed at the exact same time.

This rotted apple should not hide the good part of HFT, which is to reduce spread and inconsistencies between markets and to generate profits from this (positive) action. HFT took the place of traders, who were paid a lot for doing that stupid task.

Freakeconomics analyses how dangerous it is to walk while drunk, and found it was 8 times more dangerous than driving drunk, for the same distance. cf http://freakonomics.com/2011/12/28/the-perils-of-drunk-walki...

I tried to do a fair comparaison between comparaison between the main date implementations. The ciso8601 is really fast, 3.73 µs on my computer (MBA 2013). aniso8601, iso8601, isolate and arrow are all between 45 and 100µs. The dateutil parser is the slowest (150 µs).

  >>> ds = u'2014-01-09T21:48:00.921000+05:30'

  >>> %timeit ciso8601.parse_datetime(ds)
  100000 loops, best of 3: 3.73 µs per loop

  >>> %timeit dateutil.parser.parse(ds)
  10000 loops, best of 3: 157 µs per loop

A regex[1] can be fast, but the parsing is just a small part of the time spent.

  >>> %timeit regex_parse_datetime(ds)
  100000 loops, best of 3: 13 µs per loop

  >>> %timeit match = iso_regex.match(s)
  100000 loops, best of 3: 2.18 µs per loop

Pandas is also slow. However it is the fastest for a list of dates, just 0.43µs per date!!

  >>> %timeit pd.to_datetime(ds)
  10000 loops, best of 3: 47.9 µs per loop

  >>> l = [u'2014-01-09T21:{}:{}.921000+05:30'.format(
        ("0"+str(i%60))[-2:], ("0"+str(int(i/60)))[-2:]) 
     for i in xrange(1000)] #1000 differents dates
 
  >>> len(set(l)), len(l)
  (1000, 1000)

  >>> %timeit pd.to_datetime(l)
  1000 loops, best of 3: 437 µs per loop

NB: pandas is however very slow in ill-formed dates, like u'2014-01-09T21:00:0.921000+05:30' (just one figure for the second) (230 µs, no speedup by vectorization).

So if you care about speed and your dates are well formatted, make a vector of dates and use pandas. If you can't use it, go for ciso8601. For thomas-st: it may be possible to speed-up parsing of list of dates like Pandas do. Another nice feature would be caching.

[1]: http://pastebin.com/ppJ4dzBP

It is an interesting trick. Will I have to pay the school if I leave it early, and to reimburse any grants that I may have gotten? Do you know anyone who had done that?

I am in contact with some of them. It is better than nothing but still very hard just to have a Skype interview.

I don't want to go back to school now, I am tired of that. How about working for a US company outside of the USA, before to move within the company ?