Re: existing model, for recsys, as long as the product already exists you have some baseline available, even if it's not very good. Anything from "alphabetical order" to "random order" to "most popular" (a reasonable starting point for a lot of cases) is a baseline model.
I agree that a randomized control is extremely valuable, but more as a way to collect unbiased data than a way to validate that you're outperforming random: it's pretty difficult to do worse than random in most recommendation problems. A more palatable way to introduce some randomness is by showing a random item in a specific position with some probability, rather than showing totally random items for a given user/session. This has the advantage of not ruining the experience for an unlucky user when they get a page of things totally unrelated to their interests.
Yes, it's just really nice to be able to push a button and have it hold at the right temperature; sort of like how boiling water on the stove or in the microwave and using an electric kettle are functionally equivalent but the kettle is way more convenient.
A little desktop app that lets me upload transaction csvs from my bank and figure out how much I need to split with my partner. Mostly because I always forget to charge her for utilities or flight bookings and I hate going through the bank UI. Might also expand it with some simple subscription auditing logic.
Also, a dramatic anime intro (complete with cheesy AI generated theme song and video) starring our foster kittens. It's been interesting to learn about some of the techniques needed for consistency, how to storyboard, etc.
I don't see how their analogy is stupid. Aeropress and french press are pretty similar from an "enthusiast coffee device" perspective. Lots of room for variability in grind size, coffee choice, and specific brewing technique with both methods.
Aeropress is a brand, one I've never heard of. It fits in the Linux ecosystem (maybe as one of the Red Hat flavors?) but as an analogy it is simplified. Linux is so much bigger than that and there's everything from LFS (grow, grind and brew with tools you've sourced and put together yourself), to Android (plain old drip machine). Reducing everything that's the Linux ecosystem to a niche brand of a specific type of coffee maker is dishonest.
I use a french press myself, and never heard of Aeropress. My machines all run Debian with DWM and I never have any problems. My non-technical girlfriend is fine on Debian and doesn't really know the difference. She did mention how fast her laptop boots though.
I am too lazy and not enough of a snob to write several paragraphs on why Aeropress is objectively better and different from French Press, but it is, and I hope someone can step up and do that here.
Might have been k-nearest-neighbors rather than k-means. Knn can be used for "recommended because you bought X" or "users like you also bought X" type recommendations that relate user to user or item to item.
K-means could potentially be helpful to group together common users/items if e.g. you're memory constrained and don't want to give each user a fully unique embedding entry so that's also possible.
I have also recently been playing around with Julia coming from numpy/pandas. I found that 1.6.1 felt way smoother on startup and import than 1.5.x. Might be worth trying to upgrade if you haven't already. At the very least it's encouraging that it seems like they're making progress in that direction.
This is actually the idea behind Direct Primary Care[1]; basically the idea is that by not taking insurance the overall cost of care goes down due to the decreased overhead around the claims process.
My dad is a PCP who switched his practice over to that model several years ago. From what he's told me he prefers it a lot over the traditional model specifically because it allows him to have a subscription based model rather than charging per-visit. That means that his revenues are more consistent but also has the benefit of aligning the patients' and doctors' incentives more directly: the doctor makes more money when their patients are healthy than when they are sick so they're incentivized to do a better job.
It also meshes well with telemedicine and chat-based consultations. He was offering both of those options to his patients well before covid because it tends to save both doctor and patient a lot of time, and there's no concern over "how much do I bill for telemedicine vs a normal visit, and how do I bill insurance for it."
I agree that a randomized control is extremely valuable, but more as a way to collect unbiased data than a way to validate that you're outperforming random: it's pretty difficult to do worse than random in most recommendation problems. A more palatable way to introduce some randomness is by showing a random item in a specific position with some probability, rather than showing totally random items for a given user/session. This has the advantage of not ruining the experience for an unlucky user when they get a page of things totally unrelated to their interests.
reply