hellepardo

mlpack, a C++ machine learning library, includes xeus-cling notebooks directly on their homepage: https://www.mlpack.org/

The xeus-cling work is awesome and has made it possible to do data science prototyping in C++. There are lots of other C++ notebook examples in the examples repository: https://github.com/mlpack/examples/

> This field pushes away the Van Allen Belts, a radiation swim-floaty that surrounds Earth's middle.

"swim-floaty"? I get what they are referencing, but can we really not come up with a better term than... "swim-floaty"?

> I feel there is a bit of undercurrent of Luddite sentiment going around here.

Oh, I'm no Luddite, don't get me wrong---I'm a machine learning researcher. I have no problem with data science. The difference between all of the complex technology you pointed out inside the baby monitor and what we're talking about here is that all of that complex technology is robust and, for the most part, well-designed! The data science work here has tons of issues, and I think the vast majority of reactions are reacting to that: machine learning and data science really can be useful... but not if you apply it really badly. Someone else commented elsewhere in the thread that a simple thresholding algorithm would be just as effective---and not suffer from the myriad potential problems present with blindly applying TensorFlow because it's cool.

> the real downside of this model: it overfits on regular crying

I'm with you there, that's a big downside too, but that's not the "real" downside---there are like seven different downsides present with the data science going on here and it's hard for me to say which is the biggest issue, because they're all issues. Data science is not trivial!

So, don't get me wrong. This is a fun project and a neat little device. I don't mean to take away from it in any way.

However it is really important to consider why baby monitors are so primitive: because the cost of a false negative is huge. I didn't see any mention of this in the author's experiments (only a '>98% accuracy' note). So let's talk about this a little bit: is "accuracy" what we want? Probably not---I don't care if I get accidentally notified, but I care very much if I don't get notified when the baby is crying. So you want to weight your classifier's predictions heavily against false negatives (at the price of false positives). It would be good to make an ROC curve to characterize this behavior. More importantly though, any predictive model assumes a stationary distribution; i.e., training conditions accurately reflect test conditions. But will they in real life? What about when your neighbor's house is under construction? Can interference from chainsaws cause the model to fail to detect the baby crying? What about the dude down the street with his super loud motorcycle? What happens then? I bet the training set doesn't have situations like this.

I really, really don't want to come off like a wet blanket here. But I feel obligated to, because this is a model that directly impacts the welfare of a human, and so we should at least talk about or discuss potential drawbacks. (Again, cool weekend project, just, we need to be clear about the implications of outsourcing the decision of whether the baby is crying to a black-box model where we can't interpret what it's doing.)

Similar experience here in Georgia. It sounds like there is more to do on setup morning here, since we had to show up at 5am (and didn't have things really ready until 6:30 or so). There are so many seals to check and record; lots and lots and lots of paperwork.

In Georgia, the ballots are printed from a terminal that the voter uses and then scanned, leaving both an electronic count and a paper trail (the ballot itself is ejected from the bottom of the scanner into a sealed ballot box).

However, our scanners jammed 40 minutes into the day; after a couple hours, a technician managed to come to our precinct and opened the ballot box and revealed that a lackluster design in the ballot box caused the ballots coming out of the scanner to sometimes curl up and jam. Without any realistic solution, we just had to open the ballot box every time it jammed (supervised every time to ensure no monkey business) and push any stuck ballots out of the way of the scanner so that more could be scanned. Amusingly we had good success regularly giving the machine a good shove to dislodge any stuck ballots.

We also had problems printing receipts---in our case, we only need to print 3 from each scanner, but we ran out of scanner receipt paper. Since another precinct called us during the day looking for extra receipt paper... it wasn't available. But, we dodged a bullet, since there was just enough paper to print 2 of the 3 receipts. (1 gets posted on the door of the polling place; 2 go to the county. We wrote an apology on the receipt and only sent 1 to the county. I verified on the Secretary of State website that the votes tabulated for our precinct matched what our receipts printed. Cool to be able to double-check like that!)

I spent a while thinking about what a pollworker would need to do to illegally cast ballots. It would be a tall order indeed and would require cooperation and secrecy from everyone there, since the only way to cast a ballot is to scan it, and everyone can see the scanners at all times. I can't see it realistically happening in any precinct.

I would agree that the coronavirus could be a factor here. At the same time, I've been noticing issues since probably December or January (before the coronavirus started being a real problem), which makes it seem like maybe there are multiple issues.

Of course, I'm not actually internal to Microsoft or Github, so I have no idea and it's all opaque to me.

Yes, and so this is actually a thing that bugs me, because I use Github every day. Over the past several months, there have been numerous days in which I've had problems (`You can not comment at this time`, 500s, etc.) and no corresponding status report.

It seems like the historical uptime page paints a far rosier picture than I am actually experiencing.

Is it just me, or has Github's quality of service been continually degrading over the past several months? What is going on internally? Is this because of the Microsoft acquisition? Increased usage? An internal transition to Azure?

...is it time to move away from Github?

True, and I think that this makes the statistics somewhat misleading. The Atlanta metro area has a population of roughly 6 million and it's not very centralized.

A couple of thoughts on Minecraft, though note I am not familiar with Education Edition:

1. We played Minecraft with the specific intent of making visually appealing buildings. So at some point, when you can't see, that's not going to be fun no matter what you do...

2. Minecraft really doesn't have any accessibility whatsoever. You can scale the UI, but... high contrast mode? If you could even get that working with the base game, it's definitely not going to work with the mods we were playing with. As my friend went blind, it got harder and harder for him to deal with any zombies or anything that was moving, since it took him so long to slowly scan the screen and understand where he was. We considered trying to make mods to make things a little bit easier, but struggled with coming up with any mod that would actually improve things. :)

3. My quick glance at the API suggests that we'd end up with a situation where we were playing the game and he was... programming. That may work for some people, but I think for this group that borders too close on after-hours work...

If we are in the same place, either someone else will roll for him or he'll roll and we'll tell him what he got (and he has little choice but to trust us :)). If we aren't in the same place, he rolls virtually (did you know you can type '/roll 1d20' in Google Hangouts and it rolls?) and gets the result via a screen reader.

For a while he was DMing and he would use a screenreader to access his notes plus the rolls. We don't typically use maps or boards, but instead try to do it all with descriptions of places. It does mean that the rooms we enter all tend to be fairly simply shaped, and it's possible that each of us has pictured a slightly different room, but it all works out in the end.

Ah, sorry, you are absolutely right. When I said 'no real other option' I was missing the numerous other RPGs that are out there. I did not mean to denigrate them by omission. I meant more that we were forced away from computer games.

My friends and I play D&D because we have no real other option. We used to play Minecraft and other collaborative building games as a group, but then one in our group went fully blind. There is a complete lack of good multiplayer computer games for entirely blind players (admittedly that is quite a challenge), but D&D requires only imagination, which all of us still have. Highly recommend if you have friends with vision disabilities.

The mlpack library (http://www.mlpack.org) also has a C++ implementation of Hamerly's algorithm, as well as Elkan's and some other accelerated variants. My experience has been that while Hamerly's algorithm can be a lot faster than the unaccelerated implementation, it's not always the fastest---that depends on dataset conditions.

I don't think "emsmallen" can be said reasonably :)

Thank you for saving me from needing to write the exact same thing!

Whoa, hang on a second. SourceForge? Acceptable with new owners? Have you tried to download anything from the site recently? They may not be bundling crapware with downloads anymore but it's still an ad-laden slow-loading pile of trash unusable website. (I don't think I'm being hyperbolic there, to be honest!)

Armadillo is built on any BLAS you might like to use. (Eigen is a hand-implemented replacement for BLAS, by the way.) So I, too, am interested in what BLAS implementation is used here. OpenBLAS vs. regular LAPACK/BLAS (or ATLAS) can make quite a large difference.

> why not just get a job at Google so you can do that?

Don't want to move to California? Can't pass the interview? Salary too low in the offer? There could be lots of reasons.

I agree with your other points, but what I'm saying is not exactly that. What I'm saying is that TensorFlow is not a community-led project in the way that other open-source projects are. It's Google-led, and if you don't like it, you can go away or fork it (...and if you fork it, it's unlikely you'll get many users given the Google marketing machine).

(The same complaints have been and are being regularly made against Red Hat, and also Fedora, given that Fedora is basically "experimental" Red Hat.)

> Bigger example is how much they give away in terms of AI/ML.

You think they are publishing all these papers and open-sourcing TensorFlow for the good of the community solely?

No... if the entire ML community publishes papers with TensorFlow as code, then now Google can sell that many more TPUs and that much more access to their cloud systems to run the models...

I have always felt that, in a "typical" open-source project, I could approach the community, join it, build some rapport, and eventually help with the decision-making process (assuming I was any good and people in the community liked me).

But with Google-run (or other company-run) open-source projects, it's a completely different story. Want to be a TensorFlow maintainer? Great, sure, you can contribute, but you'll never get to make any decisions about the direction of the project itself unless you're inside Google.

To me it's a disappointing semi-abuse of the term "open source"... it doesn't mean what I think it should...

My opinion is that as long as there exists free ways to connect with people, nobody is going to pay for it. So it's tough for me to see a way to build a business around this; I don't think people will be particularly willing to pay.