Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
pps43
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
29 ms
·
271.
▲
by
pps43
9y ago
Actually ROC curves are commonly used in finance, e.g., to quantify separating power of various credit risk models.
272.
▲
by
pps43
9y ago
Unfortunately you do not usually know the loss function when developing a model. A typical example would be credit bureau developing something like FICO score for the banks to use. Banks might know the loss function, but the credit bureaus
273.
▲
by
pps43
9y ago
In this case knowing that ROC stands for Receiver Operating Characteristic wouldn't help much.
274.
▲
by
pps43
9y ago
I was trying to explain why it is more important for the satellites to quickly detect launches from land. With SLBMs you have other options. You can have your attack submarine shadowing enemy's ballistic missile submarine, or surface A
275.
▲
by
pps43
9y ago
They have 20 Romeo-class submarines, and can probably modify some of those to launch a relatively small ballistic missile.
276.
▲
by
pps43
9y ago
A year ago few people believed North Korea can build a hydrogen bomb, or an ICBM capable of reaching continental US.
277.
▲
by
pps43
9y ago
If you rely only on submarines for deterrence then you don't need satellites detecting launches in the first place. I don't think UK or France have (or need) that capability.
278.
▲
by
pps43
9y ago
How do you know North Korea has no SLBMs? They seem to be making good progress recently.
279.
▲
by
pps43
9y ago
If you detect a massive ICBM launch from Russia, you know what to do - nuke Russia. You have half an hour to launch before Russian missiles take out all silos. Now what do you do if you detect SLBM launch from somewhere in the Pacific Ocean
280.
▲
by
pps43
9y ago
Tesla autopilot is not that different from its namesake in airplanes. Autopilot does not fly the plane from JFK to LAX, it only maintains speed, altitude and heading. If anything unusual happens, autopilot disengages and drops controls back
281.
▲
by
pps43
9y ago
"Saying little" means providing only the absolutely necessary information. If you were in a self-defense situation and your attacker threw his weapon in a bush, you want the police to know that so that they can find it. If there w
282.
▲
by
pps43
9y ago
He's specializing in use of force for self-defence, not necessarily shooting. The argument goes like this: "Many people advocate that in the aftermath of a defensive use-of-force event you say literally nothing to the police excep
283.
▲
by
pps43
9y ago
This is probably too extreme. Andrew Branca recommends [1] to "say little" rather than to "say nothing". [1] https://www.reddit.com/r/CCW/comments/5sicm5/the_law_of_self...
284.
▲
by
pps43
9y ago
I suspect that North Koreans have heard of Faraday cage and shielding. The missile has no need to communicate with the outside world, it can use astro and inertial navigation systems.
285.
▲
by
pps43
9y ago
With today's computers you can get a lot done without knowing t distribution and p-values, just using bootstrap and Monte-Carlo. See 93 conversions in test and 87 in control and want to know if this is just noise? Pull 10,000 bootstrap
286.
▲
by
pps43
9y ago
The traditional way to avoid overfitting is to reduce the number of independent variables, shrink coefficients towards zero, or otherwise limit the complexity of the model. With deep neural networks the approach is different. Instead of try
287.
▲
by
pps43
9y ago
That's the explanation I often hear that I find suspicious. It's not the same muscle. Chess is a game with complete information, trading isn't. This is an important distinction, heuristics are very different. Maybe poker is b
288.
▲
by
pps43
9y ago
I can see why people with analytical minds can play chess well and create good financial models. I don't think that learning to play chess better helps one create better financial models.
289.
▲
by
pps43
9y ago
5% interest rate does not imply 5% business growth. Interest rate simply reflects the fact that people prefer $1 right now to $1 a year later. In this context banks are time machines, moving money through time (glossing over credit risk, in
290.
▲
by
pps43
9y ago
The problem with this approach is that it's hard to measure progress. I can't objectively tell if my writing has improved (if I knew I was making mistakes, I wouldn't be making them). Popularity of a blog is also not a good m
291.
▲
by
pps43
9y ago
OPM recently demonstrated how well the government can protect the data it collects.
292.
▲
by
pps43
9y ago
> that usually already have agreed upon meaning or connotation within her subspecialty. You mean, like p in statistics?
293.
▲
by
pps43
9y ago
But this is not common. If you don't know what uppercase sigma refers to, sum is a much better guess than standard deviation of multivatiate normal distribution. In this case google is probably simply case insensitive.
294.
▲
by
pps43
9y ago
Standard deviation is lowercase sigma, sum is uppercase sigma. They are not the same.
295.
▲
by
pps43
9y ago
Economies of scale. Pulling a credit score costs about $.30 and takes two seconds. How long will it take to search public court records? Besides, it's not only the bankruptcies that count. Good tradelines are also important, and those
296.
▲
by
pps43
9y ago
The problem that stun guns are designed to address does exist. There is a place in use-of-force continuum between shouting and shooting. You can't really require police officers to get into fistfights, especially when they are at a dis
297.
▲
by
pps43
9y ago
So you re-write the code from core ideas and get a different result. Now what? Did you make a mistake? Did the original author make a mistake? If you have original dataset and .Rmd file, you can then bisect both analyses to find the bug.
298.
▲
by
pps43
9y ago
Some unit tests are redundant and only waste time. Others are very useful. Do not write unit tests just because the methodology requires it. Write unit tests when the unit needs to be tested. Typical examples: * Smoke tests (does this code
299.
▲
by
pps43
9y ago
The article shows the result of this approach, and it clearly does not work well.
300.
▲
by
pps43
9y ago
Not really, because in each image parts of the watermark will be in different places. You don't even need to warp the watermark to achieve that, it's enough to randomly reposition it.
More ›