cf

Your comment expanded out is what I had hoped that blogpost would have been. Thank you for posting this. It's hard to overstate how much easier bidirectional typechecking is compared to other approaches.

My favorite starting point is actually Dunfield's lecture notes on bidirectional type checking, https://people.mpi-sws.org/~joshua/bitype.pdf

The paper you mentioned is great at showing the approach scales well to giving you all the fancy features you might want that are in modern languages.

I agree a lot of data can be represented graphically, but if you look at the literature it mostly is getting shoehorned into vector spaces. This doesn't mean people shouldn't learn about Graph Laplacians and friends, but I don't think it's an entry requirement.

I disagree about the graph theory as well. Unless you are doing things with learning on networks you won't need it.

I think a solid background in linear algebra, multivariate calculus, and convex optimization will take you really far.

Welcome to the party!

Well as another machine learning researcher, I think you can make a major impact working on problems that aren't in AI.

Now if you are a PL person interested in AI we would love to have you and your expertise!

Oh and if you want to use AI to solve PL problems like program synthesis and program completion, that's cool too.

I'm sorry you didn't feel any of the challenges had substance. They were still challenged.

This is what sklearn does for their documentation and it's fantastic!

People can express any opinion they want in public. Inside a company isn't public. Also while you do have a right to an opinion you don't have a right to it going unchallenged and questioned.

Also how can you be so sure he wasn't alt-right already. "Classic liberal" is essentially the closest you are allowed to get in public.

So one challenge I face is that many of my tabs are for web content that isn't always there. Do any of these extensions like Great Suspender, Session Buddy, etc actually let me save the webpage as it is and then let me choose to refetch it as necessary?

Do we actually give the finance industry a pass? The fact that "Wall st" is a pejorative these days is some indication that's not the case.

I don't get why if they don't mean unlimited just say up to 20TB/mo.

This book and Ross-Larson's Effective Writing series are the books that finally taught me how to write. If you write as part of your job. These books are gold.

The word is insightful.

I did not know about gradcheck. Thanks for the pointer! I have some handwritten code that does some of this for me. But essentially, yes! I want better tooling to catch my mistakes.

I wish there were more ways to unit test the gradient calculations in my code. While I try to use AD as much as possible, it isn't always feasible. Of course it isn't just gradient since if you are implementing something like Adagrad there are additional corners bugs can lurk.

I appreciate it. Anyway all my old email addresses still work if you want to get back in touch.

Hey this is cf. I had no idea this was all going on, and I appreciate you trying to shield me and other people from getting pulled into this. You dropped off Jabber and I worried if you were alright. I'm glad to see you thriving in Iceland.

Sure, but is it actually deleted? Or is this still stored somewhere in Google ready to be retrieved if needed?

The topology explanation is about what I had in mind for motivation. I think to appreciate cohomology, I need appreciate what problems in homology it makes easy to solve. To appreciate that I need have the vocabulary of algebraic topology.

One way to maybe then to describe homology is it gives you way to take about shapes and surfaces in the language of algebra.

I feel when it comes to motivation in math, I just want to know why people got so excited about a particular theorem in the first place. I'm ok with the answer "before this theorem we assumed all these different things we proved in similar ways were different. Now we know about their commonality and it allows us to borrow mathematical machinery and use it here".

For example, take something like measure theory. A reasonable motivation for measure theory to me is remembering your introductory probability class how you had to learn a probability mass function for discrete spaces and a probability density function for continuous spaces. Now these two ideas are obviously nearly the same idea. I mean the notation used is a pretty big hint. But you need measure theory to have the right concepts to describe how they are the same.

Where are all those kinds of motivation for things like topology and cohomology?

People with kids have very particular timetables they have to hit. Usually their lunch time would not be so regimented. Truthfully, though the best thing is to try it out and see the kind of people you get. I made it a point to engage with every attendee I did not immediately recognize and find why were they there and if this was their first meetup what got them there.

That seems sensible. For our meetup we assumed 50% no-show, and that 10-30% would RSVP within 24 hours of the event. We only showed the location if you RSVPed just to get better estimates, but ultimately we did the math and calibrated from there.

Consider having in during lunch hours. Especially if there is a topic relevant to work, managers will be super supportive of people going to them.

As somebody that ran a successful meetup which I grew from 0 to 1500+ members (now 2700) on meetup.com the article feels like it misses the point of the meetups.

The presenter being late, disorganized, and talking down to the audience is a pretty simple problem to fix. Screen your presenters! Don't let people who you know would give a bad talk speak. With our meetup, we would send some guidelines along with some information on the kind of audience we tend to attract. This along with front-loading the socialization addressed nearly all problems the author had.

Our meetup was and continues to be sponsored and we never felt the need to give them a plug. We thanked the companies for hosting us, encouraged people to talk to each other and that's it. None of these problems cost a penny to fix.

I have seen other comments here mention that a meetup is not about the talks that it is only networking. While, I agree that is the main benefit for going, I tried pretty hard to have cool talks with topics attendees were unlikely to run into online. I loved to spotlight people who had passion projects previously working in obscurity. I loved introducing academics to an audience of mostly industry people, and vice-versa. People would come up to me afterwards surprised to here that someone in town had solved a problem giving them pain for months! Bringing talks about things they might not see on Twitter or Hacker News.

Want to know what I actually worried about? I worried about diversity. Although 50-60% of the speakers booked were women we had trouble boosting that ratio above 10-20% for the audience. I worried about attendees with kids that couldn't make evening meetups. I worried about people who worked somewhere hard to commute from. My co-organizer actually started organizing events that occurred during lunchtime. These had just as good if not better attendance than the evening talks. So do know those can work.

I think meetups are a fantastic medium to meeting people who like similar things as you. And if none of the meetups around you are working, you can always just start one that does.

For those unfamiliar with this race, the Outside magazine article on it is a great read: https://www.outsideonline.com/1924491/60-hours-hell-story-ba...

Anil Dash actually wrote a great article about this. The short summary is you need real humans doing the moderating. All the tooling you create should be about enhancing those humans.

http://anildash.com/2011/07/if-your-websites-full-of-asshole...

My presumption is that we will interact with most computers through wireless devices. For example, the chromecast has no buttons on it, and I haven't felt limited by that.

Sure. I tend to keep my postprocessing of a model under version control. In particular, what features were most helpful for predictions.

PMML is fairly verbose and limited to a particular set of models. It's often easier to pickle the models and then keep tagged versions. I think a human readable format could be created, but since most models are just a pile of numbers it's unclear what is gained.

Yes but in this particular instance Uber recruits heavily in the Bay Area. If all workers in the Silicon Valley office refused to work and discouraged others to work for Uber it could be quite effective.