dswalter

I used to use Spark SQL for this purpose, but I've switched. I now use Spark for when I want to transform data. But when I'm writing ad-hoc data exploration/investigation queries, PrestoDB is my jam; it's what it's designed for. Parquet as the data storage method makes both of these quite workable.

I've been using PrestoDB for a few months now, and I'm deeply in love. It's such a well-designed piece of technology. The query execution engine is a tremendous boon to anyone with inconvenient-sized data. And it does most (all?) of ANSI-SQL!

The wait is over! (my deepest apologies)

On the positive side, the corrosive powers of the dollar from your example appear to be only logarithmic.

I've seen UNIX tools that were more robustly designed chatbots.

That's a [edit:perfectly fine] example to choose for a Stanford MOOC, since Coursera was started based on the success of Andrew Ng's machine learning course.

edit: People below (and above) are more knowledgeable than I am.

If the only major impact of current deep learning methods on culture was to squeeze out an additional 1-5% of performance on every task set in front of it, the fact that it has made large-scale speech recognition and image recognition good enough for public use, it would be enough to call it substantial progress.

But I think one of the major successful things of the deep learning renaissance has been the ability to embed different 'worlds' into the same vector space (French and English, text and music, images and text, etc.). By co-locating these different worlds, we gain the ability to perform more accurate search, to create images from text like in the "Generative Adversarial Text to Image Synthesis" paper, and a wide variety of other multi-modal tasks. We can multi-embed almost anything, even things that are nonsensical. You want to make a music-to-font translation system? Or a sneaker-to-handwriting generator? Gather the training data, and the world is now your oyster. The impact of deep learning as differentiable task pipelines has only begun to scratch the surface of what is possible.

After all the draws, it was nice to see some actual wins.

For those less familiar with chess, the final game was kind of fun. Carlsen's final play was to sacrifice his queen to draw the king into a trap, when Karjakin was one free move from putting Carlsen into an inextricably losing situation.

A major suspicion at the time was that other Android OEMs were concerned about Google's production of their own hardware and potential ensuing favoritism. To this day, Google has branded phones (Pixel, etc.), but they aren't made by Google.

As an aside, my Moto X is still such a well-designed product.

Since 1000 miles/year is not enough to pay for actual use for almost anyone, this move effectively means new model Teslas have to pay per Supercharge visit. Non-free Supercharging should open the quick-charge market up to much broader competitive market forces.

This SMBC comic makes the same point on survivorship bias. http://www.smbc-comics.com/comic/path-of-a-hero

I'm guessing you meant to link this instead. https://github.com/Microsoft/LightGBM/wiki/Experiments#compa...

Excellent writeup. The post makes use of the OP's deep learning library Keras, which is utterly fantastic. If you want to learn more about deep learning, I'd recommend getting started with Keras on top of Theano or Tensorflow; the API is elegant and keeps pretty up-to-date on modern tweaks to structure/initializations, etc.

I wish there were a clearer delineation between developer downloads on PyPI and automated builds, since it seems there's been a bit of a cultural sea change over the past year or so in favor of Py3. I'd love to know which version humans are downloading.

Modularization aside (and it's terrific to have in the library), what additional functionality are you most excited about with this release?

I would posit that Snapchat and Facebook are working so hard on videos because video ads have a much higher payout in the current ecosystem. I don't know that it's necessarily because of efficacy.

I've seen a decent amount of complaining about the difficulty of supporting a kafka infrastructure. I'd be interested in thoughts from people who have used Heron as to how it is running it in production.

It's interesting to see that as further work is done on spark (and I'm pleased they're actually improving the system), it behaves more and more like a database.

It was a nice touch.

That convivial atmosphere is one of the things that originally set Lyft apart. More recently, they've tried to figure out how to balance that with people who prefer silence. http://thehub.lyft.com/blog/2015/7/14/giving-betterrides-gre...

In addition, with the attention fantasy sports has been getting, his competition is probably stepping it up a bit. It may be easier to beat everyone's uncles than to best teams of people who are also using statistics/machine learning/sports analytics.

That is a heart-wrenching encounter. I admire the author's cool head under pressure and emotional honesty and awareness.

Are You a Human | Data-Engineer/Backend Engineer, Senior Data Scientist, Intern positions| Detroit, MI | Full-time | ONSITE.

I'm a data scientist here, and the question we work on is a fundamentally interesting one: Which users on the internet are human? As a bonus, the people are smart, interesting, and pleasant to work with.

apply here: areyouahuman.com/careers

Author here: whether the hyperparameter tuning is done automatically or manually is not as important for what I was trying to say here. But yes, any of {grid-search, random-search, bayesian-optimization, etc} is likely to be more effective than manually tuning to squeeze out those last ounces of performance.

OP here: I probably should have made it clearer that the using the same train-test splits is verboten.

Your second and third paragraphs are also exactly correct. I attempted to make those points in the post, but you've done so more effectively here.

While dual-axis plots can be useful to have, there are some strong arguments from the data visualization community about how they are sub-optimal. https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&c...

It's not really directly comparable. This is a computational framework which you can use to implement a variety of ML algorithms (or in general numerical computation), so the interface to this is lower-level than Azure ML.

If Azure ML is a bunch of premade sandcastle molds, TensorFlow is a more accurate, faster way to pour sand. You make the mold.

Atlanta's a major air travel hub, so extra-cheap flights are partially a result of that. If the poster above you has parents that are off the beaten path, it's very likely those $49 deals don't exist for the flight she or he needs.

You bring up a really good point. Medium's default layout scheme highlights content over authorship, so you don't get smacked in the face by the fact that this is written by Jay Carney.

Are You a Human | http://areyouahuman.com | ONSITE strongly preferred | Detroit, MI USA

Looking for Data Scientists, Front-End and Back-End Engineers

We’re dedicated to making the digital world better for real humans. Automated programs, called bots, account for 61% of overall internet traffic today — and as the world becomes more digital, this will only increase. Ultimately, bots make it more difficult for real humans to get what they need from the Internet, and more difficult for companies who serve humans to provide high quality content and services cost-effectively. We find and verify humans by analyzing natural user behavior across millions of websites.

All humans interested in joining the team, please visit http://areyouahuman.com/careers to apply.

--------------------------------

I work here as a data scientist, and the central task of verifying humans opens up a fun bag of machine learning and engineering challenges. It's also an amicable, smart group of humans to work with.