HNHacker News
TopNewBestAskShowJobs

cityhall

231 karma · joined March 15, 2012

submissionscomments
cityhall··on Python For Finance: Algorithmic Trading
It sounds like day trading for script kiddies. There's no way to develop or execute a sound strategy with that level of access.

It's like a slot machine tournament. No one has an edge but some percentage of people will do well enough over a finite period to convince them to put real money behind their lucky algorithm.

cityhall··on Don't use deep learning when your data isn't that big
I'd say the benefits come from the information content of the features or lack thereof. When you have uninformative features like pixel colors or word identities, there's nothing for traditional methods to work with. You have to start with feature engineering and pruning before decision trees or linear classifiers have a chance.

Most of the wins under the "deep learning" umbrella come from extracting meaning from homogenous features like "the pixel at x-2,y+1 has red=123" or "the word at n+1 is 'king'". That's why we see latent variable embeddings like word2vec come from the DL world even though they're not deep.

When you want to include highly informative features in a deep network, it's often better to feed them into a separate logistic model, as shown in the Tensorflow wide-deep tutorials.

cityhall··on How Bayesian Inference Works (2016)
This is handled by choosing a smoother, higher entropy prior. If you have uncertainty about your prior then basic factorization tells us it's equivalent to integrating over your various priors with respect to the probability you assign each of them.
cityhall··on AI Is Getting Better at Detecting Handguns
They seem to be spotting visible guns, not concealed ones, and the focus is on guns being brandished instead of holstered.

This is preliminary work but it wouldn't be useful to have an alarm go off every time another police officer or security guard comes into view with a gun on his hip.

cityhall··on Maine Is Drowning in Lobsters
If we make the same argument in favor of taxi cartels, people see the flaw right away. The system may encourage conservation but that's only one of many side effects and we're glossing over the costs.
cityhall··on Why Are Economists Giving Piketty the Cold Shoulder?
Normalizing the concept of a wealth tax is productive even if it's unimplementable. The right has used the same approach by having think tanks promote flat consumption taxes and zero corporate tax rates.

If nothing else the absence of a theoretically desirable wealth tax serves as a justification for progressive income taxes and higher capital gains taxes.

cityhall··on Show HN: Howmanypeoplearearound – Use wifi to calculate number of people around
I assumed they had deals with the cellular companies to get aggregate location data to estimate traffic.
cityhall··on Analyzing 10 years of startup news with Machine Learning
Articles can't be treated as data yet except with respect to very simple questions.
cityhall··on Facebook and Google were conned out of $100M in phishing scheme
The real professionals are the ones who don't talk about it.
cityhall··on New Evidence Suggests Humans Arrived in the Americas Earlier Than Thought
Do bones maintain whatever properties they're exploiting thousands of years after the animal's death? I'd expect they'd have different wear or fracture patterns if they were very old when worked.

You'd also expect a mixture of ages in close proximity. Even if all the bones being processed were old at the time, they presumably would have a mixture of 100kya, 50 kya, etc. bones jumbled together.

cityhall··on Tech companies should create ‘farm teams’ to close the talent deficit
Employees aren't a manufactured product. The claim is the supply is inelastic in the short term. You can outbid your competitors for talented developers but a rise in average salaries won't make weak developers better or get lawyers to take a code bootcamp. The only short term ways to increase supply are immigration and outsourcing.
cityhall··on Wealth isn’t created at the top, it is merely devoured there
The problem is we don't tax wealth directly, only when it's transferred. A large portion of government spending is to protect the value of capital, but we fund it largely with taxes on incomes.
cityhall··on Show HN: Aeneas – a Python audio/text aligner
There's nothing new about this, it's how speech recognition training data has been generated for a long time. Whether you can align a script will depend on how accurate it is and how expressive your models are for generating spoken/surface form alternatives for the ways things like dates are verbalized, which look different in text. If more than one person is speaking at the same time the results will be terrible.
cityhall··on The Uber Bombshell About to Drop
They're a decade late and have no engineering culture. All they could do was throw money at the problem. If they weren't hyper aggressive they would have looked to partner with someone more experienced or to play vendors off each other.
cityhall··on The Uber Bombshell About to Drop
What you want is competition, not for one entity to control the market regardless of its organization.
cityhall··on After CIA leak, Intel Security releases detection tool for EFI rootkits
There's more than one country interested in pulling off these attacks. The US can say, "Do us this favor and we won't look too hard at your tax avoidance schemes," but China or Russia might have an easier time planting a few employees.
cityhall··on ANTLR Mega Tutorial
This was my experience too. Did you find any standard cost-based robust parsers or are we still stuck rolling our own?
cityhall··on New kind of recurrent neural network using attention
How do GRU cells do on these problems?
cityhall··on World’s richest doctor gave away millions, then steered the cash to his company
In other words he wants founders who will do shady things to make him a profit but not to the point where he loses plausible deniability.
cityhall··on TensorFlow 1.0 Released
It's worth pointing out Tensorflow is basically Google's clone of Theano, including a lot of the same design decisions. They've improved some things but it's not like Google handed us the secret to fire here. It's just a good implementation of the same things a lot of people have been working on for years.
cityhall··on TensorFlow 1.0 Released
This is good, the argument order and axis handling was inconsistent.
cityhall··on Ultrasound Ages Liquor Two Years in Just Three Days
What ultrasound thing should I search for? A quick attempt turned up strange skincare and aromatherapy products.
cityhall··on Silicon Valley Hedge Fund Takes on Wall Street with AI Trader
Not all algorithmic trading is high-frequency where you need to shave microseconds to beat someone else's order. If you're predicting actual price movements instead of order book fluctuations there's no reason to use hardware.
cityhall··on What Vizio was doing behind the TV screen
The problem is virtually every company is on the list for one reason or another. Sony's rootkit for example.
cityhall··on Russians Engineer a Slot Machine Cheat that Casinos Can't Fix
Your bankroll needs to be much higher than that to withstand the expected variance. Your edge shouldn't really depend on the size of the bets.
cityhall··on Russians Engineer a Slot Machine Cheat that Casinos Can't Fix
With correct play slots have some of the lowest house edges per dollar bet. Sports betting has a fairly high edge, 4.5% on normal bets and much higher on exotics. The difference is you make a few sports bets and walk away, while with slots you keep cycling money through until you lose it.
cityhall··on Russians Engineer a Slot Machine Cheat that Casinos Can't Fix
The irony here is it's only illegal to beat the games because the casinos figured out how to rig the democratic process to get favorable laws passed. If we put it to a vote, what would the public think was more unethical, using a mechanical device to count cards or using lobbyists to make sure you can't lose?
cityhall··on Copper is key in burning fat
The same thing can happen with copper IUDs.
cityhall··on How to use Deep Learning when you have Limited Data
I think his point is no one can tell you from theory which regularization methods to apply to a particular problem to get the best results. You need expert knowledge, experience, and hyperparameter tuning.
cityhall··on Ask HN: Why aren't there middle class cars shaped like Ferraris?
That explains why you don't see track-inspired designs but not why you don't see cheap clones of Bentleys and higher end Mercedes.

Hyundai actually seemed to break with convention a few years ago and make their cheap cars resemble more expensive competitors, presumably because they didn't have anything in that market to devalue.

← PreviousPage 2 of 3Next →