brilee

22 ms·

211.
by brilee 8y ago
It also includes the exponentiation step before the vector normalizations. There are connections to statistical mechanics here, where the relative energy population numbers are proportional to the softmax of the energy levels divided by tem
212.
by brilee 8y ago
Yeah - this kind of tool is ripe for crypto mining abuse.
213.
by brilee 8y ago
Unsure why this is downvoted. When averaging results over a whole city, you've eliminated most sources of noise. But if you average neighborhood by neighborhood, then you have 20x smaller samples, and on top of that, 20x more chances t
214.
AutoGraph converts Python into TensorFlow graphs(medium.com)
14 points by brilee 8y ago|2 comments
215.
by brilee 8y ago
Agreed that the extraterrestrial origin hypothesis doesn't actually explain anything. It just merely adds a factor to the Drake equation of how likely it is for life to get started.
216.
by brilee 8y ago
There's an open hypothesis that life may have evolved elsewhere in the universe and then brought to Earth by some means. DNA would probably have diverged by now, but who knows?
217.
A Deep Dive into Monte Carlo Tree Search(moderndescartes.com)
150 points by brilee 8y ago|6 comments
218.
by brilee 8y ago
The birthday paradox arises because with N people, there are N^2 possible pairings of people for birthday collisions, and thus the number of people required for a collision scales as sqrt(D), an unexpected result. But here, there aren'
219.
by brilee 8y ago
You're right that matplotlib has an unpleasant API - I've never had a good experience with matplotlib, but I've used Altair for a few data viz tasks and love it. Matplotlib intentionally took on the API that Matlab had create
220.
Visualizing TensorFlow's streaming shufflers(moderndescartes.com)
1 points by brilee 9y ago|0 comments
221.
by brilee 9y ago
Not everyone gets the luxury of traveling around the world.
222.
by brilee 9y ago
IIRC the difficulty with titanium was not in the production, but rather the difficulty in machining the metal afterwards.
223.
by brilee 9y ago
"If a line begins at character 0, that is a list of files whose timestamps should be checked" The list of files would be the declared dependencies.
224.
by brilee 9y ago
In about half of the examples, the methods are more user-friendly wrappers around underlying libraries - great! love it! In the other half of the examples, the niceness of the API is in fact due to the user-friendly design of the underlying
225.
by brilee 9y ago
This is a nice wrapper around requests, pyquery https://github.com/gawel/pyquery/ and parse https://github.com/r1chardj0n3s/parse of which only requests is Kenneth Reitz. Let's give cred
226.
by brilee 9y ago
I'm not sure how that article applies; it discusses the hazards of using a PRNG. I think the actual objection is that it just increases the number of samples required, but doesn't fix the problem.
227.
by brilee 9y ago
The last name Choi is pronounced Chae by Koreans. Don't see why it's so scary to you; Richard and Dick and Rich are in the same boat.
228.
by brilee 9y ago
Here's one of the stronger models we've gone through. https://github.com/tensorflow/minigo/releases/tag/v199
229.
by brilee 9y ago
You'd have to write a Chess implementation that very carefully respects all possible game-ending pathways, a translation of a chess board into an array that a NN could understand, and a schema by which to flatten the array of all possi
230.
by brilee 9y ago
We're working on making all the data (models, selfplay SGFs) available to download - you could download an older generation.
231.
by brilee 9y ago
About a month using K80s. We have something like a million games played.
232.
by brilee 9y ago
With a lot of elbow grease, mayyyyybe. We're currently using 128 filters, 20 residual layers, whereas LZ is using a smaller network size. Our networks would have had to define their batchnorm layers exactly the same way for the model f
233.
by brilee 9y ago
We've put it on KGS as "somebot", and it can play at somewhere from 3-5d level, although amusingly it still can't do ladders, and almost always plays 3-3. Our strength seems to have plateaued recently, and we suspect it&
234.
Minigo: An open-source implementation of the AlphaGo Zero algorithm(github.com)
206 points by brilee 9y ago|29 comments
235.
by brilee 9y ago
This is one of the most distasteful "well actually" comments I've read in a while.
236.
by brilee 9y ago
Excellent. I'd been wanting to implement something like this but these guys beat me to the punch. Most existing research on NNs in chemistry uses the SMILES string representation and feeds it into a recurrent neural network. The proble
237.
AlphaGo Zero: an analysis(moderndescartes.com)
2 points by brilee 9y ago|0 comments
238.
by brilee 9y ago
Did anyone else find it bizarre that they quoted "grams of protein per ounce/cup" of food? America seems to have slowly adopted the metric system, but it's not what I expected to look like...
239.
by brilee 9y ago
From the article: > Then, in 2014, her pharmacist told her that her insurance plan would cover only the brand-name drug, which cost her family some $50 more a month than the generic. If she paid for the generic herself, it would not have
240.
by brilee 9y ago
Vast majority of these are not actually oddities. A few stuck out at me, though: - since True/False are aliased to 1/0, you cannot have both 1 and True as dictionary keys. This could plausibly happen in real code. - there's n

More ›