Hacker Read top | best | new | newcomments | leaders | about | bookmarklet login

It always surprised me to see them reporting results to a 10th of a percent precision (e.g. "2.3%"). I'd assumed that the best they could do on an individual was a couple of orders of magnitude less precise (e.g. 10%, 20%)


sort by: page size:

92% is an awful lot of precision.

In the article they claim 93.5% accuracy.

The method may be good, but the 86% is very suspicious. For this kind of thing I'd be happy if they can show more than a half accuracy, or 3/4, but 86% looks too precise to be a real number.

I find it difficult to square a 10-40% chance of success with "fairly good accuracy".

numbers suggest around 2% accuracy (7.1)

It's more likely he just rounded a number of results and rounded to the first decimal place to display it better, not to suggest some level of accuracy.


90% accuracy, presumably on their test set, seems less impressive than it may sound.

How accurate are the results?

It's just a thought experiment. If the accuracy is -+n%, then lets say "2*n% worse".

Reducing error by 25% from a 96.5% baseline gives you their stated 97.25% accuracy. About 0.75% fewer errors. Still amazing, but less impressive than the abstract makes it sound.

There's no false precision. There's a whopping +/- 40% error bar! What can be said is that 26% is at least less precise on average. You've added a completely arbitrary 0.32 because of some obsession with decimals.

70-90% accurately only, iirc, although the article’s method is also only 90%

Rather looks like 10% accuracy in this case

It seems a little premature to proclaim "Near-Perfect Accuracy" on such a small sample size. It sounds to me like they may have overfit their model.

Wow. Those accuracy number are crazy, I had no idea they would be so low (<5%).

I'd be more concerned about the phony precision. Not 250%, not 270%, no -- 273%.

That's true. 82% isn't what we'd consider very "accurate", so it's something to take with a grain of salt - there are always exceptions.

90% accurate sounds impressive, and it is, but its still 100% incorrect almost always.

If even the experimental approach is only 90% accurate, how do they know which 90% is accurate?

What do you mean with accuracy here? Usually 50% accuracy means cointoss, meaning 20% accuracy is equal to 80% accuracy, which is better than the article's 78% and not that far from 90%.
next

Legal | privacy