This is a nice project. I think it would be great to add references used for the implementations and some tests that demonstrate they return what is expected (or perhaps the same result of sklearn maybe).

Those are great suggestions. I will look into adding that.

I think it's the best part. The definition is left for the student to learn.

Their data scientists will now be building features like "tried to watch a movie from a VPN IP" for their churn models.

And another note: there are also people who simply do not allow comcast to filter their traffic and therefore always have a VPN. These people despite their efforts to watch netflix content approved for their own country, would be blocked.

I tried it with my favorite, Alex Webb. It didn't break 5/10 and received solid #mess and #boring tags on all of the photos. Fun none the less.

The main finding from my perspective is that we can score and rank friendship likeliness and see signal in a hold out dataset.

I can almost hear Facebook clamoring to offer free genotyping.

The dating world is already experimenting. (e.g. http://www.genepartner.com/)

I see the optimization problem they face. Promote non-personal content to generate short term revenue versus promote personal content to minimize long-term churn. It looks like their data scientists and business analysts have identified significant risk to pursuing the former and advocate the latter, for now.

In other words this has nothing to do with a return to claimed core values, but a shift in strategy to remain afloat for as long as possible.

The real problem is a post IPO Facebook and its obligation to make money. Intimacy isn't profitable and there is no money in sharing personal photos. Publisher content has money but FB risks becoming a feed of useless updates and viral fluff. Within this space, text is a poor revenue generator but video has the highest CPM so we are seeing a major shift to video. FB's $50mn payouts to BuzzFeed and the like shows their commitment to viral junk. The watermelon live video gimmick isn't long term sustainable. FB is losing the plot it seems. There was a time when people used to post pics of things they ate and do FB checkins everywhere they went. It got people on board but other than on-boarding it proved to be just a fad.

I imagine the long term strategy will be to selectively optimize feeds based on user prefs. Not everyone values intimacy and some people prefer BuzzFeed cruft on a daily basis.

Unfortunately intimacy is hard to measure while sharing and content generation makes for easy KPIs for exec to focus on. Focusing on these will treat the symptoms but not the cause. In fact if sharing and recirculation is a KPI then FB will increasingly gravitate toward Twitter.

However this is all backwards. Monetizing the users is a piss poor business model. It fucks up the UX and alienates the users. FB should move to other avenues. Personalized search or ecommerce or premium SMB accounts should be the bigger focus.

Aside, their move into messaging reminds me of Microsoft's play when they leveraged Hotmail users into MSN and virtually overnight, uprooted ICQ/AOL.

> Intimacy isn't profitable and there is no money in sharing personal photos.

Facebook makes money as long as users continue using its services. It doesn't matter much what they are doing as long as they keep coming back.

Exactly, sharing personal photos is insanely profitable if it keeps people on your website. You need to have a core value prop and just like TV it needs to be the primary user experience. Facebook is just readjusting for that after initially ramping up the money firehose.

It shows Facebook has a long term focus which you would expect from a public company.

We are in agreement. Intimacy is a loss leader.

And then, back in reality, FB is minting money.

Advertising has been a huge business for the last 100 years (or more) and it is all going to move online.

The more people deploy adblockers, the more FB makes: (statistically) everyone on mobile access it (and other FB-owned brands) via the app, so they get the revenue of the money that would go to the open web.

I don't use adblockers so I'm not 100% sure how they work, but don't mobile ones work at some system level to prevent other apps (e.g. Chrome) from showing ads? Why wouldn't they also target other popular apps with ads like Facebook?

The kind of mobile ad blockers that work on the system level require root permissions afaiaa. Meanwhile, uBlock Origin is available on Firefox for Android, but being a Firefox extension, does not have the possibility of blocking traffic in other apps.

The amount of people who have rooted Android devices are probably far lower than the likely-also-low number of people using in-browser ad blockers on Android. For iOS devices, the number of people with jailbroken devices I would guess to be lower in relative percentage than on Android. I don't know if even just in-browser adblocking is possible on iOS at all without jailbreaking.

Anyway, point is, while ad blocking might be relatively wide-spread on the desktop among all people (why else would media bother to spend valuable developer time implementing ad blocker blockers?), in-app ad blocking probably is much less common and that's what fb benefits from.

There are a variety of adblocking technologies. The spectrum goes from the simplest which just block domains, to ones which block substrings of URLs and DOM elements to ones actively stopping "please stop adblocking" notices.

In general AdBlocking works very well on the open internet with the browser of your choice. However, they're basically non-functional for things happening in-app over HTTPS connections. So, with the majority of FB mobile traffic happening over the dedicated FB app - they are largely immune to blocking AND have a much better set of demographic targeting features.

No, they are all implemented as browser addons.

Everyone props up "Growth Hacking" when what we need is product managers who have a feeling of how to please customers on the long term.

That's also the base salary. Yearly bonus can easily be 100% of the salary. Partners at decent firms are of course clocking ~2M or more.

Can you give some concrete examples of the benefits of wrangling in python over R and a general sense of how much time this would save a user?

I really would be interested because I started in Python and R and now use R exclusively due to manipulation. R (dplyr) really is the best manipulation system have ever used. Perhaps people try to do loops in R and other non-R like ways and give up. R is fundamentally a Functional Language and people try to band a square into a round hole and give up?

This is consistent with my impression of how sales work. You sell a feature, product, capability, etc that won't be ready for at least 6 months (for example). Then you kindly encourage an engineer to build it.

I agree. The title should read 'Where Can a Ph.D. Take You? Forward to More Research, Usually'

Which is the entire point getting a PhD in the first place. More research. Forever.

I (along with many others) cringed at the phrasing of postdoctoral degree. A post-doc is not a degree. The article's comparison of a post-doc to law school for liberal arts majors was equally egregious. Sloppy writing.

Now, to be fair to their point. What they are observing is a cultural phenomenon in academia which is to stay in academia until you can't stay any more. Industry is for those who fail at academic research, whether that failure occurs at the post-doc stage or pre-tenure stage. This is in contrast to the sentiments of the real world (the perspective of the article author) which wonders why these "students" take so long to get a real job.