The video was interesting. Seems like a nice way to start a shopping search if you have a picture with something you want where the look matters. Eg, cars, furniture. etc.
Not directly an answer, but similar what if thinking got me wondering: wouldn't it be cool to be able to use postgres much like how you use sqlite with python? I implemented this idea as a pip-installable python package, https://github.com/orm011/pgserver, and your feedback would be great :) I use it for my projects.
I have not compared it.
Have you had issues in this front?
If you are asking in human terms, its instantaneous to start, it simply creates a few files etc and starts a process. If you were using Sqlite, there is no extra associated process. But the processes are simply waiting for your input, so normally this is not really extra work.
In benchmark terms, the way I'm using it startup time is not a huge issue, instead being able to use postgres is helpful (has extensions and you know you can eventually move to a hosted postgres if you want)
If startup really matters, the startup can be amortized (if multiple processes want to access the same file, they get a handle to the same already started process, so only the first one would have waited for startup).
In terms of ongoing resource consumption, I can easily have a bunch of separate servers running for different files (it depends on your workload how much work will happen behind the scenes.) I'm not sure how much extra work postgres needs to do vs sqlite beyond what is kind of inherent for a workload (eg index creation, saving a bunch of data etc).
Hm, not exhaustive but I think these are potentially useful to you:
The deeplearning.ai math basics for deep learning, seems self-contained.
MiniTorch repo (implement your own tiny torch) seems also helpful to understand what goes on during training.
MinGPT repo (to understand a basic version of GPT model structure)
Dive into deep learning (textbook avail online, more focused on practical DL)
I have been waiting for weeks, and am still waiting, to get access to Llama2 (released a month+ ago), and access to this model goes through the same form, so I'm not very hopeful. Are you getting it from other methods?
Does anyone know of good tools to practice tone-listening and recognition for cantonese? or tools for drilling jyutping more generally,
I've used https://chaaklau.github.io/cantorocks/, but I would like to also try others, as the audio is not great sometimes and I'd like to listen to different voices.
When studying Mandarin I found that getting the phonetics right early on really helped me, it can be done in a relatively short time and its hugely motivating.
If you have basic Mandarin knowledge, you could try Duolingo's Cantonese course for Mandarin speakers. I think it is helpful to get familiar with how words are pronounced and what expressions people use in vernacular Cantonese so far. As with the Mandarin course, there is little explanation, you kind of have to complement it eg with some drilling for tones.
I really like this idea, though I haven't tried it yet.
Spotify has lyrics for many songs, but for characters the first hurdle is you cannot really sing along even with the lyrics unless you know them.
Is there some easy way one change the font used by Spotify to be this font somehow?
I'm not sure the author will ever read this, but I am also a PhD student, have been one for far too long, and in the past have struggled with similar `time management' issues, though I'm not sure that's really the problem.
As time passes, especially when things go wrong, projects and writing become very emotional. Having memories of looking at something a few years ago and looking at it again a few years later to try to resubmit it tends to trigger guilt, anger etc in me. I've found some strategies to help me, because I'm close to finishing and think the marginal benefit/marginal cost makes sense (benefit is large, finish your phd in 2 years, 1 year, 6 months, depends on where you are).
One tool I use to help me work even when I'm not feeling it is Focusmate (google it). I've found that sometimes, working out my issues is useful , but sometimes the emotions are overpowering and no amount of working them out will make me not regret something. That used to make me stop. Nowadays if I use Focusmate, I tend to think less of such `large picture' issues and instead work on just a little bit, and get a little done. then after a day, a get more done, and so on. I do sometimes get upset and avoid it altogether, but I'm definitely much better at `time management' this way. Also, I don't want to leave a focusmate in the middle of a session, i just need to stick around 25 minutes, and the feelings can pass and i can even enjoy some of the craft of the specific thing I'm doing. I hope it helps someone.
3Blue1Brown videos seem like a good resource to use along any book. My experience as a math major (in the distant past) is that the kind of visualization the author shows you is also something you want to imitate in your head when you are learning new concepts. I find things I learned in this level tended to stick in my head 10+ years later, other stuff less.
I should mention focusing on doing a few interesting problems, rather than many not so interesting ones, is also one way to help yourself understand more deeply.
Lots of easy problems is a good way to build up muscle memory, though. IMO the brute-force method of, say, Saxon Math really makes sense for things like basic elementary school algebra and probably intro calculus, where the student is sort of learning the math equivalent of how to walk. Not sure where the switch over ought to be, though.
we tried "Saxon math", Singapore math dimensions, and Beast Academy.
And my impression was that Saxon Math was the worst.
What I mean by worst is that it just make you practice an algorithm by doing lot of repetition but doesn't force you to have a deep understanding or problem solving skill.
Saxon math worked out for me, although we didn’t shop around as far as I remember, so maybe Singapore would have worked fine as well.
My experience is that I didn’t really feel like I was memorizing an algorithm. Because the problem set includes assignments from all of the old sets, it is hard to memorize all of the algorithms. So you instead memorize the different moves that are allowed and have a general idea of what types of moves might be useful.
I dunno. I went on to do engineery stuff as an undergrad rather than pure math stuff, it seems like a good match because engineering problems are also often in the “no need to be super clever, just don’t mess up” vein, so it could be just a lucky match. This is what I mean by muscle memory — I’ll use the famous names theorems when necessary but sometimes you just need to bash the math until the thing you want is on that side of the equal sign and the other stuff is on the other side.
I think anything that results in
1) actually reading some textbook
2) actually working through problems for a couple hours a week
will compare well to the typical US math education pretty well anyway.
