Hi @zeagle, sorry to hijack the thread (didn’t see a way to DM you)
I'm a PhD student working on a new lupus diagnostic blood test approach [1]. Hoping to steer the project towards true clinical needs.
I'd love to ask for your feedback as a technologist + rheumatologist on a few lupus + RA diagnostic directions we're considering. Would they actually be useful in your practice?
Would you be open to a quick chat? My email is maximz@stanford.edu.
Love this. I appreciate your building on ripgrep versus my own bulky lucene-based approach a while back (https://github.com/maximz/sift), and that you don’t require pre-indexing but build up a cache as you go.
I may have what you're looking for: I've been working on exactly this problem, because I grew tired of the headaches from maintaining a "snowflake server" (http://martinfowler.com/bliki/SnowflakeServer.html), which I configured once in the past and would be very difficult to reproduce.
Here's a super-simple Docker host stack I put together that you can install on a cloud machine fairly easily. It runs Docker containers, does automatic backups, and has monitoring and alerting built-in. I've run around 5 webapps, split between 10 containers, on this infrastructure for about 6 months with significantly fewer headaches. Monitoring alerts go straight into a personal Slack channel.
The implementation is deliberately low-level -- you'll be running bare docker commands -- because the goal is to have a trivial infrastructure and to learn Docker in the process. There's no reliance on the "magic" in big wrappers like Kubernetes. Because there's nothing fancy in here, it's simple to understand and use.
My write-up is rough around the edges right now, but eventually this will turn into a simple blog post with an Ansible playbook attached. Here's a rough draft of how to set it up manually: https://github.com/maximz/docker-host/blob/master/public_doc...
There's an nginx reverse proxy fronting the Docker containers. You can do a zero-downtime deploy pretty easily. Happy to answer any questions if they arise.
I saw David Donoho give this talk live in September at Princeton's Tukey Centennial conference -- fantastic, and well worth a read. IIRC, gives a good history of data analysis, how to think about the different definitions of and roles for data science, and an introduction to Tukey's work.
For more on the history of data science, here are references from a similar talk by Chris Wiggins: http://bitly.com/icerm
I think this is because of different risk preferences, which make the conclusion obvious.
These teens have higher risk preferences. Entrepreneurship is a high-risk, high-reward endeavour.
Since we count only the successful ones, those who take higher risks and are successful naturally are more successful than those who take smaller risks.
My major gripe is that some formatting is completely hidden, making it impossible to deal with.
I've often tried to copy-paste from an email thread to a new compose window, which looks like it works. But then the resulting email has a line break after each word!
The clear-formatting button doesn't reveal this, nor does plain-text mode - this has led to a number of embarrassing emails.
I wasn't a CS major while there, but I did take a few CS classes. Those classes, in particular, are top notch, and a lot of time was clearly spent on at least the intro classes (the first 3-4 classes you may take). The classes are very well polished, the support staff is welcoming and teaches well, assignments are always really fascinating, and the courses as a whole are very put together. It's part of the reason that the CS department size more than doubled in the last few years!
I'm currently at Berkeley EECS. It's certainly a very good program, and they put a ton of effort into undergraduate education. For example, I was recently talking to people at the ParLab about how they developed a rather popular class about parallel programming, so undergraduate education is certainly something of a priority.
Ultimately, it really depends on what you're interested in. From my perspective, Berkeley seems very good for systems stuff: parallelism, distributed systems, databases or even computer architecture. The undergraduate CS classes are all rather good and mostly interesting. (Except for the stupid Java/OOP/data structures course--that was a big waste of time, but you can probably just skip it.)
However, the classes really aren't the most important thing. I think there are two things that trump this. One is simply location: Berkeley is very close to a ton of startups. I've met some incredibly smart people running exceptional companies, and even worked at a couple during the year. It's a great resource that would be hard to duplicate anywhere outside of the Bay Area. Even if you don't plan to spend much time working at startups, it's still really cool to occasionally pop down to SF and hang out with hackers; the community around here is really great.
The other aspect that's more important than classes is research. Berkeley has a very project-driven approach to research where people decide on applications first and tailor their research to that. Some projects are really full-stack: for example, at the ParLab (where I'm doing some undergraduate research), there was a demo where everything including the hardware, the OS and the high-level algorithm design all stemmed from active ParLab research, all integrated together at the very end. If you like that sort of thing, it's very neat. The other big lab I know about--the Amp lab--operates similarly.
It's also going to depend on what particular subjects you like. From my perspective, Berkeley is great for all sorts of systems stuff: architecture, distributed systems, parallelism and so on. It's really a great atmosphere for that sort of work, with a ton of cool people to talk to.
Of course, this does mean that most of the stuff that's going on is annoyingly practical. Relatively few people seem to be working on neat theory, and most of that is in complexity. So if you want to study cool theory just because it's pretty, Berkeley might not be ideal.
I've found myself inexorably drawn to PL theory--types, categories, algebras and so on--and Berkeley is really no place for that :(. It's driving me to distraction. So if you're interested in something along those lines, I'm willing to bet that Princeton is quite a bit better. And I really do think PL theory is the coolest field, partly because of how general it--it's applicable literally everywhere in CS and then spills over into math or even physics--but mostly because of how pretty it is.
Also, if you like functional programming--another thing I'm very interested in, unsurprisingly--then Berkeley is also the wrong place. Most people seem to know literally nothing about it beyond its being somehow related to functions and a bit wonky, and I've only met a few people who know much beyond that.
Anyhow, I hope that helps you get a good idea of how Berkeley is like. I'm very biased, but I think my biases cancel out: on the one hand, I go here and it's great; on the other, the things I actually like are sorely under-represented and that really annoys me.
I just want to back this up by saying that I've been exposed a little to some AMP lab projects (Spark, Shark), and they're pretty cutting-edge for the industry. That lab seems to be focused on tackling the same problems industry is actively working on, which is rare at my university. If you're interested in distributed, parallel computing in the vein of MapReduce, I would second Berkeley.
I'm a PhD student working on a new lupus diagnostic blood test approach [1]. Hoping to steer the project towards true clinical needs.
I'd love to ask for your feedback as a technologist + rheumatologist on a few lupus + RA diagnostic directions we're considering. Would they actually be useful in your practice?
Would you be open to a quick chat? My email is maximz@stanford.edu.
Many thanks!
[1] https://www.biorxiv.org/content/10.1101/2022.04.26.489314v5