tbenst

22 ms·

181.
by tbenst 8y ago
Wonder how this compares to single shot with an SVM or nearest neighbor. 76% on MNIST is frankly embarrassing
182.
by tbenst 8y ago
Lot of misinformation in the marketing. This is not a checking nor savings account. Checking and savings accounts are FDIC insured which gurantees $1 in / $1 out. This is a brokerage account, and if the way robinhood invests the cash g
183.
by tbenst 8y ago
10x faster is a bit exaggerated; maybe 2-3x faster than typical desktop FFF and still slower than HP or Carbon
184.
by tbenst 8y ago
The new JIT is very interesting. Anyone know if this is for inference only or also for training?
185.
by tbenst 8y ago
I think this is because when you do a “market order” on Robinhood it’s actually placed as a limit order 5% above the price. Just set your own limit order next time
186.
by tbenst 8y ago
I was introduced to probabilistic programming by Frank Wood and Brooks Paige at DARPA’s PPAML program, and they are excellent teachers. Highly recommend this resource! Anglican is a great language embedded in Clojure so this is particularly
187.
by tbenst 8y ago
I looked at Elm a couple years back and decided on Purescript instead: http://www.purescript.org/ Has a much better FFI story with JavaScript than Elm, and has a more powerful type system (higher kinder types, etc). Row pol
188.
by tbenst 8y ago
By reference to HM, I mean to illustrate that Rust has fundamental similarities to the typed lambda calculus, which has been done more formally by Rust Belt. Additionally, traits in rust are highly similar to type classes in, say, Haskell.
189.
by tbenst 8y ago
I’m not sure where your perception is coming from. Rust and Scala are more functional than they are OOP. Rust for example uses Hindley-Milner type inference, putting it squarely in the functional camp. And Scala is highly functional as well
190.
by tbenst 9y ago
This article misses the point for the Scientist in data science. HPC will not support docker, Singularity seems to be the winner: http://singularity.lbl.gov/
191.
by tbenst 9y ago
The challenge isn’t just the hardware. Software in neuroscience is in an exceptionally sorry state. As I write, I have a job running to infer the number of neurons and their spike trains from a recording done with 4096 electrodes. Sampling
192.
by tbenst 9y ago
I would recommend looking at Singularity as well: http://singularity.lbl.gov/ . Docker is not supported in the HPC world but Singularity is showing rapid adoption
193.
by tbenst 9y ago
How flexible is this compared to Church/Venture, Webppl, Anglican, etc? Does it support recursively-defined generative processes? Edit: nvm, Noah Goodman is behind this, who created Webppl. This looks super flexible and awesome, congra
194.
by tbenst 9y ago
There's also a code base for GHC based on this work called Subhask. It's technically Haskell but replaces so much of the functor/monad hierarchy that it's effectively a different language. https://github.com&#
195.
by tbenst 9y ago
I don't think the comment limits the meaning of abandonment, but rather gives an interpretation for a specific circumstance. Here's some umpires debating this rule: http://www.umpire.org/vb/showthread.php?s=2
196.
by tbenst 9y ago
Julia however does not use Hindley-Milner[1]. I'm under the impression that HM is not compatible with multiple dispatch, which would be very unfortunate as typeclasses can serve the same purpose while maintaining compatibility with HM.
197.
by tbenst 9y ago
Misleading reporting. The author truncated the actual MLB rule 7.08a, which reads in full: 7.08 Any runner is out when— (a) (1) He runs more than three feet away from his base path to avoid being tagged unless his action is t
198.
by tbenst 9y ago
And, four years later, they still have not fulfilled their kickstarter promise of the Ghost Dashboard. This was the most requested feature on their public trello roadmap before the card was deleted one day. Over the years there have been ma
199.
by tbenst 9y ago
I looked at OCaml a few years back for scientific computing, but was turned off by the global lock GC for multithreading (similar to python). Anyone know if this has been changed yet? Edit: doesn't look like this has changed, but there
200.
by tbenst 9y ago
I ran at an elite D1 collegiate program. The vast majority of runners will get injured at 100 miles per week, assuming a reasonable pace no slower than 7 min / mile. Slower than that is not useful training for Olympic events. Our top r
201.
by tbenst 9y ago
FYI, you can simply clone an overleaf document using git, and work offline, pushing changes after: https://www.overleaf.com/blog/195-new-collaborate-online-and...
202.
by tbenst 10y ago
Just a heads up that church has now been replaced by webppl: http://webppl.org .
203.
by tbenst 10y ago
Unclear what a "laser 3D printer" means--possibly selective laser sintering, but that doesn't produce carbon monoxide to my knowledge. Possibly a homegrown machine. Tragic reminder of the importance of proper workspaces.
204.
by tbenst 10y ago
Yes Stan is awesome! The main difference between something like Stan and "next gen" languages like Anglican and Webppl, is there are basically no restrictions in where you use a distribution. Nested inference, probabilistic recur
205.
by tbenst 10y ago
You don't need to know what that is to use the languages :). In essence, functions always take one argument plus a "continuation," which takes the value that would have otherwise returned. This is called a trampoline. It
206.
by tbenst 10y ago
Anglican for high performance (Clojure host, strong multithreading support) and WebPPL for ease of integration (JavaScript client side or server side). Both are implemented using continuation passing style trampolines.
207.
by tbenst 10y ago
I attended DARPA's Probabilistic Programming for Advancing Machine Learning (PPAML) summer school. The aim of Probabilistic Programming languages (PPL) is to abstract away the act of Bayesian inference into modular engines such that sw
208.
by tbenst 10y ago
There is another plausible explanation: there are many people, especially photographers, near the turn for lanes 1-4 but none for 5-8. The start has people on both sides. The rest of the deck is barren. Even after 15 years of competitive sw
209.
CephFS production ready with release of Jewel(ceph.com)
6 points by tbenst 10y ago|0 comments
210.
by tbenst 11y ago
Matt is an incredible human being. I reached out to Matt (as a stranger) on behalf of a relative with a rare genetic disorder and was humbled by his support and generosity. Matt is a true mensch and a hero in rare disease diagnosis. What wo

More ›