tbenst

16 ms·

121.
by tbenst 6y ago
I have great respect for Stephen Diehl and love his writing. But I must respectfully disagree. I think it’s fantastic that Haskell is seeing more paying jobs and corporate sponsored development thanks to crypto, and indeed Haskell can help
122.
by tbenst 6y ago
HDF5 has no support for dataframes/tables, although projects have been built ontop of it like PyTables, but these are non-standard and don’t work across languages.
123.
by tbenst 6y ago
I’m also very confused about the relationship of Arrow’s stability guarantees and that of the on-disk feather. Can we safely switch from parquet to feather for long-term data storage?
124.
by tbenst 6y ago
“In other words, information travels from one location to another through the same number of synapses.” —-Yaniv Assaf This statement is outrageous, and an example of how some scientists gain the ire of their colleagues by exaggerating the
125.
by tbenst 6y ago
Yes, their statements should be about “large scale processing blocks,“ to make claims to the public about the number of synapses between neurons being conserved is really unfortunate and cannot be assessed in their data
126.
by tbenst 6y ago
Thanks for the thoughtful reply. I’m highly concerned that there are artifacts baked in. I’ll readily admit that I’m a systems neuroscientist and am skeptical of DWI in general. I appreciate that they included Supp. Fig 13, with a compariso
127.
by tbenst 6y ago
This is outrageously misleading. To make a claim about the number of synapses between neurons based on MRI data is completely unwarranted. Voxel size (single volumetric pixel) in MRI is approximately 1mm, while synapse size is way less than
128.
by tbenst 6y ago
MRI is ~5 orders of magnitude less precise than EM. Not even close to cellular resolution, let along single axon resolution. You can only see axon tracts, where thousands of axons may make up one pixel.
129.
by tbenst 6y ago
I’m not sure this even qualifies as science. Doesn’t even compare to a group of schooled children. There may indeed be positive outcomes associated with unschooling but this gives us little useful information for assessing this.
130.
by tbenst 6y ago
I see a huge bump in numpy performance with AVX-512, to the point where I wouldn’t buy a cpu without it. I don’t know that I understand the critique—-is it because these instructions are Intel specific and not on AMD? Seems obviously useful
131.
by tbenst 6y ago
It’s not react, but author of this post also developed a hooks library for Purescript Halogen: https://github.com/thomashoneyman/purescript-halogen-hooks
132.
by tbenst 6y ago
I ran the test 15 months ago using example code from Metalhead vs PyTorch examples repo. Unfortunately my test consisted of staring at nvidia-smi, so don’t have code handy. I believe I benchmarked Resnet. Edit: I also had an issue with VGG
133.
by tbenst 6y ago
Or, they’ll follow the path of entrepreneurship from makerspaces in 3D printing and found 2000 nearly identical companies all making an FDM printer. Synthetic biology is fundamentally different from software and much harder to build a busin
134.
by tbenst 6y ago
> Genuinely curious about your data here. I invest a little in the space so have seen some pitches at IndieBio, YCombinator, and talked with a bunch of companies. There are indeed some really great founders with biology background as you
135.
by tbenst 6y ago
SynBio in the 2020s smells a bit like green tech in the 2000s. I see a ton of founders and funders with zero biology background thinking that they have the vision and can just hire biology expertise. Biology is ruthlessly unforgiving and co
136.
by tbenst 6y ago
Ah, I hadn’t appreciated this. Thanks for making & sharing your code!
137.
by tbenst 6y ago
First of all this is very cool. Dunno if author is on here, but I’m curious why both Flux and Knet are used rather than just one of them (Flux seems the most Julianic?). Also, is this really faster than PyTorch/TF? Last time I benchmar
138.
by tbenst 6y ago
I think Purescript has a joyous handling of records. Anyone know what the closest equivalent is for Haskell?
139.
by tbenst 6y ago
Using nix, you typically build from scratch and only include binaries that are needed in the Docker container. It’s quite elegant, and uses the nix cache, too, so you aren’t dependent on order of layers https://nixos.org/nix
140.
by tbenst 6y ago
I’m most concerned about the new reporting from Boyd, that according to gold standard ELISA retest of positive samples, just under half came back negative, indicating a possible false positive. Of course, the samples going through freeze th
141.
by tbenst 6y ago
Not quite, but close. “tf32” is 18 bits, but with the same 10 bits of exponent that fp32 has. It’s the range of fp32 with the precision of fp16. It’s a shame to see such unoriginality in new number representations. I’d much rather see Posit
142.
by tbenst 6y ago
Important to remember the half-precision tensorcore misrepresentations where the 8x improvement over fp32 claimed on Imagenet with tensorcores (V100) was actually only 1.2-2x [1,2]. Furthermore, there are major precision issues with network
143.
by tbenst 6y ago
They’ve also been sitting inside a hospital for who knows how long. They cite another paper as saying ICU prevalence is 30-40%, but that’s an entirely different patient population. There is no control
144.
by tbenst 6y ago
We all gain more insight by discussing the study’s or analysis’s flaws, and improving future studies. I learned nothing of interest scientifically from the discussion of Bogan. Attack the substance, not the author.
145.
by tbenst 6y ago
This strikes me as classic, Reddit-style conspiracy theory. I have no idea who Andrew Bogan is, but he didn’t play a major role in the execution of the study. I’m sure on the paper’s published the contribution section will specify what he d
146.
by tbenst 6y ago
I think the criticism in this thread and elsewhere is a bit too harsh. It’s by no means a perfect study, nor the last word, but hopefully will motivate further studies. I volunteered on this study and talked with hundreds of the participant
147.
by tbenst 6y ago
Huh, cool! I’d love to run Illustrator/Photoshop on my NixOS box. I presume you use looking-glass-client from nixpkgs. Any other parts of the config you can share/recommend?
148.
by tbenst 6y ago
I used a Ansible for years before switching to NixOS. The problem with ansible is it is an imperative configuration tool that makes efforts to be declarative, but is not fully. Most runs fail to be declarative, eg even something as basic as
149.
by tbenst 7y ago
That would be wonderful if you’re able to! Also doubles as a good intro to Jax :). Please feel free to tweet at me (@tbenst) or email [same username at stanford dot edu] if you do get around to it.
150.
by tbenst 7y ago
Has anyone benchmarked Jax? Curious how it compares to PyTorch for nontrivial networks, say ResNet.

More ›