Use whatever you want on your one off personal projects but use something more non-data science friendly if you ever want your model to run directly in a production workflow.
Productionizing R models is quite painful. The normal way is to just rewrite it not in R.
I've soured a lot on directly productionizing data science code. It's normally an unmaintainable mess.
If you write it in R and then rewrite it in C (better: rewrite it in English with the R as helpful annotations, then have someone else rewrite it in C), at least there is some chance you've thought about the abstractions and operations that are actually necessary for your problem.
Been awhile since I've worked at one but it is usually grounded in trying to achieve 100% MS usage.
It is rarish to find a partial MS shop. Most of this is how hard MS makes it to use other tools. Even in 2025 they have good interop with external tools hamstrung.
Example: SQL Servers JDBC driver will convert an entire table's of data from ASCII to UTF and a full table scan instead of convertering your UTF bind to ASCII and using the ASCII based index. This doesn't break interop but does make it painful to code and one more reason to just use .Net.
There is no way a reasonable person would not deploy to Linux and postgres for cost reasons alone. No one wants to pay Microsoft or Oracle money for databases, operating systems or frameworks nowadays.
All my .Net web apps are now deployed to Linux and Sqlite. Good riddance to Windows Server and IIS (which was dogshit from day one). With the tiny memory profile of .Net 10 it's crazy how small a VM you need to get good performance.
I've been using unfi protect/capture (I self hosted capture for a long time) for years and have never had a forgotten adoption any they almost never go down. I do have everything on UPS now but I never saw the issue before that either.
That said I've only used the wired bullet cams so maybe other models are not so nice.
Really the only downside I've seen is about 5ish years ago, all the bullet cams I bought would die after about .75 -> 3 years. All died with the same issue and I had 100% failure rate with any bought during that time frame. Ubiquiti replaced the ones that died during the warranty period but most died just after that expired.
The ones bought before or after that have been great so the issue was solved but I have a nice stack of dead ones that would work great as fake cameras, especially as their IR leds still light up.
Small companies really magnify the extreems. Good ones are really great but bad ones are extra bad. Sadly, they are also nimble enough to switch between them, at least in one direction.
Not only the extremes, also the speed: good employers can turn into bad employers (has the opposite ever happened? I'd love to learn of an example!), but big companies at least have some inertia while it happens. There's probably even some "Sun" still left, all those years after the Oracle takeover. Compare this to what happened at Komoot.
I do not have an example of the opposite, but I can echo your comment.
I was the first US employee of an Indian consulting startup. I was their engagement lead for a marquee account for the first 4 years and while I do not take all the credit, my management and I grew the account from 1 person to 250 by the time I left. What did I get in return? A 10% reduction in salary from my previous job, almost no pay hikes (there were some) for 4 years, a whole lot of "we are family" talk, and zero stock. Of course I was naive and did not have things in writing, but I still believe they owe me 3% of an 80 M exit price because that's what they verbally told me. But no, good employers turned into bad employers very quickly.
Of course there is a lot more to the story, I had my own faults, but I am not naming anyone and I am not publishing my story here. That life is over, I am not fighting that battle, this was 15-20 years back and I finally did move on and do other stuff.
But that 3% after a decade or more of (well managed) growth would have been awesome.
I have seen the opposite happen, but I'm fairly confident that few people that felt the pendulum swing from good to bad stick around long enough to feel the upward swing.
I'm OOTL, what happened at Komoot..? Vaguely interested because I'd considered applying there for a role a year or so back but never went through with it...
Founder(s) sold to Bendinggspoons who specialize in the kind of takeover where the buyer stops all development and tries to keep the money inflow from a service running with minimal maintenance crew. Evernote is the most famous example I think.
(paywalled, but I think it's the definitive aftermath writeup, as opposed to all the older news that stop at speculating about layoffs that had not happened yet)
It's more that big companies spend a ton of money hiring HR and developing process to ensure they regress to the mean whereas small ones can't afford or don't yet need that overhead so they don't have that force acting upon them and can go whichever way.
And not true, at least for the newest version. V4 has touch sensors for adjusting the temps on the side of the mattress.
I do own of these and while I hate the price, the subscription, the fact that it didn't work for an hour last night due to the internet being down (first time ever really) but there really isn't a better option. I love the temp control and would use anyone else if they had a valid competitor, but sadly there isn't one (or at least wasn't when I bought mine). The alternative is to not have temp control which is pretty amazing.
They were actually changing the deck in way that survives shuffling, not just looking at the differences.
They were using the offset on the printing as a way to tell orientation of the card. Since auto shufflers never rotate the cards, any rotation they added would persist allowing a way to tell good from bad cards in future hands.
Yes that is why I mentioned it was nearly impossible to replicate. The final optimized method involved a lot of social engineering, which required to have very high standing in the casinos. She had to request, under the guise of superstition, a specific setup with a specific style of dealer, who never changed decks, and to be authorized to call out certain cards as "lucky" which the dealer would flip themselves.
It also required deep pockets, as just playing the shoe enough to sort it could take a few hours of regular gambling. That's the crazy thing, this elaborate setup just got them a few % edge on the house which they milked relentlessly.
Use whatever you want on your one off personal projects but use something more non-data science friendly if you ever want your model to run directly in a production workflow.
Productionizing R models is quite painful. The normal way is to just rewrite it not in R.