chillee

21 ms·

241.
by chillee 8y ago
VSCode uses ripgrep internally to do the cross project searching.
242.
by chillee 8y ago
I think this is an interesting read on the topic: http://slatestarcodex.com/2014/09/30/i-can-tolerate-anything...
243.
by chillee 8y ago
Yeah this is one of the "standard" uses for matrix multiplication. Another one is calculating the N-th fibonacci number in logarithmic time (or any linear recurrence).
244.
by chillee 8y ago
It basically is path compression. You still need to maintain failure links, but at every step, when you step up your parents to find all matches, you jump directly to leaves.
245.
by chillee 8y ago
Ehh, I understand your point in general, but I think it's significantly more nuanced than that. Take the car mechanic. His incentives don't completely align with yours. Taking a benign example, his goals are to keep your car in go
246.
by chillee 8y ago
Some other type features that OCaml has that Haskell doesn't are polymorphic variants and structural polymorphism (row types).
247.
by chillee 8y ago
I only took a quick skim, but it doesn't look like they do the optimization to aho-corasick where you store connections directly to "leaf" nodes (ie: nodes where you can have a finished match). If I'm right, that would p
248.
by chillee 8y ago
Cmon, wtf? Some of the criticisms here just aren't even close to valid. He spends a half of the blog post criticizing them spending 100 GPUs on the Imagenet classification experiment, > So they trained it using 100 GPUs (100 GPUs de
249.
by chillee 8y ago
I'm really curious how much this actually causes Amazon's revenue to drop. You can still buy things and search, but consumers are probably going to be put off by the home page message. EDIT: And now it's back up.
250.
by chillee 8y ago
That shouldn't be an issue. Check the README. This is a common issue people run into on Macs.
251.
by chillee 8y ago
Since that user didn't specify them, what would you view as these warts?
252.
by chillee 8y ago
See social shaming here[1]. >Social shaming also isn’t an argument. It’s a demand for listeners to place someone outside the boundary of people who deserve to be heard; to classify them as so repugnant that arguing with them is only dign
253.
by chillee 8y ago
At Facebook, you could alter code even after somebody had given the OK for code review. I know some people specifically kept some small commits open after being approved, just so they could quickly make changes without needing approval if t
254.
by chillee 8y ago
The Foobar coding challenge seems to work really for Google, for whatever reason. Go to Google and search "python list comprehension" in a bunch of different tabs. Make sure it's exactly that. If you have an extra 's&#x2
255.
by chillee 8y ago
I think the problem is easier viewed as just a standard graph traversal (although you can view graph traversals as DP...) https://ideone.com/rU5COm
256.
by chillee 8y ago
Could you provide a citation on "the numbers hold up all the way to PhDs"? I found https://www.aps.org/programs/education/statistics/fraction-p... Vs https://www.aps.org/programs&#x2
257.
by chillee 8y ago
Well, I think you're kind of misrepresenting his point as "competitive programming doesn't improve your programming abilities". It's not that competitive programming correlates poorly with job performance; it's
258.
by chillee 8y ago
I'm not sure that's true. I think it's pretty tough for somebody who doesn't understand a field to assert that understanding the field deeper is not necessary. You'll see many non-programmers assert that it's n
259.
by chillee 9y ago
Well, ML doesn't explicitly use it for immutable data. It's just how you declare variables. It does happen that most variables are immutable though.
260.
by chillee 9y ago
Well, not really. Training a model is one such task. It's hard to train a network but easy to verify that it has good performance (training vs inference). The real problem here, I believe, and I've seen this idea pop up several ti
261.
by chillee 9y ago
There is indeed work at FB being done towards that last part.
262.
by chillee 9y ago
The heavy lifting done in neural networks is all offloaded to C++ code and the GPU. Very little of the computation time is spent in python. ASICs will definitely be very helpful, and there's currently a bit of a rush to develop them. G
263.
by chillee 9y ago
The things that have most improved training neural networks since you left were: 1. Smart (Xavier/He) initialization 2. ReLU activations 3. Batch normalization 4. Residual connections The GPUs and dataset size were definitely very impo
264.
by chillee 9y ago
I think it's pretty clear that neural networks are developing concepts/abstractions of the object it's classifying. Check https://distill.pub/2017/feature-visualization/appendix/ I think some o
265.
by chillee 9y ago
Do you have a source that cites the += example? I can't seem to find it on the page you linked.
266.
by chillee 9y ago
Assuming you're talking about adversarial examples in general (a one pixel attack would definitely be stopped by a median filter), yes. Median filters, gaussian filters, gaussian noise, all don't provide significant barriers again
267.
by chillee 9y ago
If you're interested in more experiments, I think it would be interesting to take a look at the batch norm variance/bias parameters after training a network, as well as the variances of networks trained with/without batch nor
268.
by chillee 9y ago
I was talking about the graphs here: https://i.imgur.com/M6P71aC.jpg I missed that he's not storing activations for those graphs, he's storing activations+batch norm. See my edit.
269.
by chillee 9y ago
I'm aware. He's using ReLU in this case though.
270.
by chillee 9y ago
A couple of things: 1. How can activations be below 0 in a ReLU network? It seems like: > h_out = z * (z > 0) is the line that you're using to store activations. How can that be below 0 for any values of z? EDIT: Whups, missed th

More ›