rsfern

17 ms·

211.
by rsfern 4y ago
Symbolics.jl looks really cool, I haven’t used it much but I definitely want to I think I was looking into it at the start of this project, but it looks like it’s symbolic solver can still only handle linear equations, and I’m working with
212.
by rsfern 4y ago
Oh, so you have to hook it up to a js plotting library then? Still seems cool, I knew SymPy could emit LaTeX but I didn’t know it could do js too
213.
by rsfern 4y ago
The interactive js visualizations are super cool, I didn’t know SymPy could do that Lately I’ve been using SymPy and Pluto.jl together, it’s been great for deriving results and making interactive visualizations for numerical code at the sam
214.
by rsfern 4y ago
I think that bit is saying the chemists who synthesized the compound were desensitized to it, it’s right after a passage where they were ostracized and sprayed with deodorant at a restaurant during lunch
215.
by rsfern 4y ago
Here you go https://www.researchgate.net/publication/368171693_A_room_te... I don’t know about Argonne, but at NIST all our papers are supposed to have full text up on PubMed within I think a year. I try to post to arx
216.
by rsfern 4y ago
If your output range is bounded like that then you probably want a sigmoid or tanh activation (with some shifting and scaling maybe) on the output layer. But the hidden layers can still use ReLU without issue
217.
by rsfern 4y ago
For what it’s worth, I usually default to swish activations, which seem to be popular in my corner of graph neural nets (materials and chemistry). Performance is about the same as ReLU, and I like swish because it doesn’t have a hard discon
218.
by rsfern 4y ago
Cool! To be honest I don’t really use dvc much, but the project version control features are what really interest me. I like how data pipelines help align versioned artifacts like model checkpoints and visualizations with the datasets and c
219.
by rsfern 4y ago
Oxen seems more like git (with GitHub integration (Oxenhub)) for ML datasets, where DVC is a bit bit more like make (with S3, LFS, etc integration) for ML datasets. It seems like Oxen has finer granularity version control and diff capabilit
220.
by rsfern 4y ago
There’s some interesting discussion about how to potentially improve the design of the search space. Plus they had to manually simplify the final optimization algorithm, so it’s not like they’ve cut themselves completely out of the loop, it
221.
by rsfern 4y ago
Personally I didn’t find your tone really warrants an apology, and anyway the search feature is pretty unobtrusive :) Also for what it’s worth I kind of disagree with the top level poster, search can come later after there’s enough discussi
222.
by rsfern 4y ago
HN has a search bar at the bottom of the page
223.
by rsfern 4y ago
Cool idea! One of the great things about HN is the high quality discussion. Maybe you can think of some ways to bootstrap that? Like ask submitters to leave a micro review saying why the result is cool or something they learned or some way
224.
by rsfern 4y ago
They’ll just use Google’s (or whoever’s) chatbot, but pipe it through another LLM that detects and removes ads, or highlights product placement or something
225.
by rsfern 4y ago
The point isn’t that the government doesn’t fund the research or set the big picture priorities, it’s that funding decisions aren’t really made by some random bureaucrat in isolation. In my experience NSF/DOE/DOD program managers
226.
by rsfern 4y ago
Really cool project! I’m curious about the factorization of the optimizer parameters, do you have a functional way of thinking about / tuning each of the terms? Why write `'bias_decay': .85 * 4.8e-4 * batchsize/bias_
227.
by rsfern 4y ago
Hybrid can solve this, at our group meetings (physical science, not SWE) we’ll sometimes all have laptops up at the conference table to collaborate on a document or whatever. We have a large physical screen that I think is better than shari
228.
by rsfern 4y ago
A lot of social workers are not government employees, there are a bunch of private counseling practices, NGOs, and hospital systems and the like that employ social workers as well. (I’m only familiar with the situation in the US) But I thin
229.
by rsfern 4y ago
i think there are only a couple exceptions to the Fortran/C++ rule. DFTK.jl is the only one I know written in Julia, and there’s GPAW and pySCF in python
230.
by rsfern 4y ago
This looks like a great and really comprehensive resource! Karpathy’s recipe for training neural nets is one of the only similar documents I’ve come across http://karpathy.github.io/2019/04/25/recipe/ Th
231.
by rsfern 4y ago
True, but I still think chatGPT is just a tool, maybe more powerful than grammarly, but I think it’s generally still fairly low level in terms of Bloom’s taxonomy. Authorship is supposed to be for intellectual contribution. If LLMs get go
232.
by rsfern 4y ago
If anything it should go in the acknowledgments section, not the author list. Would you list grammarly as a coauthor?
233.
by rsfern 4y ago
Pydantic can do validation. There’s also the related Typer, which builds command line interfaces from the type annotations on your entry point
234.
by rsfern 4y ago
Maybe not practical for your workflow, but making interactive plots in Pluto.jl is so much nicer than hacking around in matplotlib If it was easier to use virtual environments with PyCall I think I’d abandon matplotlib entirely.
235.
by rsfern 4y ago
Some of us federal scientists use LaTeX! Have to use Word for internal documents (like proposals) that go to management though…
236.
by rsfern 4y ago
That’s cool, is that from https://proceedings.mlr.press/v97/feldman19a.html ? I hadn’t seen that result before, definitely interested in related work
237.
by rsfern 4y ago
BNL has a bunch of software engineers doing cool data infrastructure stuff. The current project lead for matplotlib works there! https://www.bnl.gov/staff/tcaswell
238.
by rsfern 4y ago
I got a fourth round, bivalent pfizer/BioNTech flavor, a couple weeks before I went to a conference in Boston. Didn’t get Covid there (a bunch of people I was with did), but there’s no way to know how much that was just luck. I had a
239.
by rsfern 4y ago
Discussion last month when the preprint was released: https://news.ycombinator.com/item?id=33823170
240.
by rsfern 4y ago
What is missing? Is this confined to the software environment, or also covering like data accessibility and workflow documentation? Asking as a scientist interested in reproducible research. Julia has been so much better than python at leas

More ›