HNHacker News
TopNewBestAskShowJobs

Bill_Dimm

616 karma · joined June 11, 2012

The founder and CEO of Hot Neuron LLC. He developed the algorithms for conceptual clustering, near-dupe detection, and predictive coding used in the company's Clustify software. He has experience in the development and application of mathematical models to solve real-world problems in theoretical physics, mathematical finance, information retrieval, and e-discovery.
submissionscomments
Bill_Dimm··on A Word Is Worth a Thousand Vectors
Good observation -- I missed that (obviously). They seem to be using data from the word2vec project, so I would guess that it is intentional rather than a lack of cleaning.
Bill_Dimm··on A Word Is Worth a Thousand Vectors
It's not removed from the list -- it is second from the bottom. madsravn's question is a good one.
Bill_Dimm··on Gigaom Closes Down
I've been saying exactly this for quite a while. It is so incredibly stupid to replace the ads when the back button is hit. The probability that I hit the back button to click one of the ads is quite high, and the probability that I'll want to click one of the new ads that replace the one I was trying to get to is virtually zero.
Bill_Dimm··on Thunderbird Usage Continues to Grow
I just did some tests using "Edit > Find > Search Messages" to search (I'm not sure what asutherland means by the "global search mechanism") and based on a few tests searching for "contains" matches on the body text here are my conclusions:

1) If you use quotes, it is looking for the quotes in the text, so I'm surprised you got any matches at all.

2) If you don't use quotes, it is still looking for an exact match (except that it's not case sensitive), so searching for "open house" without quotes should not match unless the words are adjacent to each other.

3) It doesn't seem to pay attention to token boundaries, so a search for "house" (without quotes) matches "warehouse"

So, it is a little unintuitive (and really should have a help button that explains exactly what it does), but it seems to work for me. I don't use IMAP, though, and the comments by qznc about IMAP [1] seem like they may explain why you are seeing what you are seeing if you use IMAP.

[1] https://news.ycombinator.com/item?id=9124667

Edit: I just noticed e15ctr0n's link to info on the global search -- I don't have that enabled (probably disabled it to save disk space a long time ago) so if that's where you are seeing problems, that explains why I never noticed.

Bill_Dimm··on Thunderbird Usage Continues to Grow
Really? I don't remember having any major problems with search recently. Search is a few orders of magnitude better than the junk mail classification, at least for me. I don't know what algorithms they use for junk classification, but here are some observations:

1) I mark 100% of the emails from a particular sender or with a particular subject line as junk, yet the junk classifier still puts emails from that sender or with that subject in my inbox. This is the low-hanging fruit -- the stuff that is so obvious that it is spam that a human can figure it out without looking at the message body at all, yet Thunderbird can't seem to get it right no matter how much you train it. I can only guess that they are placing too much weight on the body content (which is easily padded with hidden text to con the algorithm) when a bad sender/subject should trump whatever is in the body.

2) Emails from people that I've emailed in the past sometimes end up in the junk folder. I thought they were supposed to be whitelisted automatically (by automatically being appended to some special section of your address book).

3) I've implemented message filters to detect certain words like our product name and move the incoming message to a high-priority folder, yet messages that should match the filter still end up in the junk folder on rare occasions (and, yes, I have it configured to run the filter before junk classification is applied -- I'm assuming that junk classification should not be applied if a filter moves the message out of the inbox).

4) It sometimes seems to just stop learning, so I have to reset and retrain the junk classifier or it gets to the point where the only emails that it recognizes as being junk are the ones in a foreign language.

Bill_Dimm··on Thunderbird Usage Continues to Grow
Any idea why? On the surface, that seems completely insane.
Bill_Dimm··on “A spambot barfed its post recipe on my blog”
I agree. When I follow kuschku's link (starting at the 900th result), there is no next page link at the bottom, and if I try hacking the "start" value in the URL to go even one document farther I get no results and a "Sorry, Google does not serve more than 1000 results for any query. (You asked for results starting from 901.)" below the search box.
Bill_Dimm··on “A spambot barfed its post recipe on my blog”
Seems like a reasonable hypothesis, but you can easily prove that it's wrong by doing a search for a single word, so there is no step (3).

I did a search for "hackerspace" (no quotes) and it claims "About 588,000 results." Paging through the results, I eventually got "In order to show you the most relevant results, we have omitted some entries very similar to the 390 already displayed." Even with omitted entries included, there only seems to be 832 results.

Bill_Dimm··on The Hidden Psychology of Wearing Glasses
Counterpoint: If you are extremely nearsighted, contact lenses can give you much better vision than glasses. If I look at the red digits on an alarm clock from across the room while wearing glasses, and I tilt my head, I can see the numbers shift relative to the position of the alarm clock (think about light going through a prism and the colors separating). I had an old pair of glasses where the numbers would shift completely off of the clock and appear to be floating in the air, but my newer ones aren't that bad (I guess the lens material is better, or the lenses are just smaller so you can't look through them at such an extreme angle).
Bill_Dimm··on LibreOffice Beta Available for Android
Not to diminish the accomplishment, but the press release title says "...make LibreOffice Beta available..." whereas it is apparently a LibreOffice viewer, not full-blown LibreOffice.
Bill_Dimm··on Internal Versus External BLOBs in SQLite
> How old is this benchmark?

It says they used SQLite version 3.7.8. Since 3.7.8 came out on September 19, 2011 and 3.7.9 came out on November 1, 2011, it was probably in that range.

Bill_Dimm··on Powerful, highly stealthy Linux trojan may have infected victims for years
Ah, it's getting stuck on /var/spool/postfix/public/pickup

Rerunning with "-D skip"

Bill_Dimm··on Powerful, highly stealthy Linux trojan may have infected victims for years
This seems to get stuck for me after a few minutes (grep stops taking up CPU cycles). I thought maybe it was getting stuck trying to read something it shouldn't, but lsof gives no clues.
Bill_Dimm··on Robots, Not Humans, Fake 23% of Web Video Ad Views, Study Finds
I think a clearer way to put it is this: The 23% number was measured across networks displaying ads on many sites. Some of those sites employ botnets to generate fake views, giving them much more than 23% fraudulent views, while other sites that don't do such things have 0% fraudulent views. The amount the advertiser is willing to pay is presumably set by their experience across the whole network, so if they were willing to pay $X per ad without fraud, they will pay 0.77 * $X due to fraud, which means that the honest websites get less than they deserve (based on the value they deliver) while the fraudulent websites get more than they deserve. So, the fraudulent websites are stealing from the honest ones, not stealing from the advertisers.
Bill_Dimm··on Deep Neural Networks Are Easily Fooled
OK, I found where Hinton claimed that[1], but it would be nice to have a better reference than a Reddit comment if anyone knows of one.

[1] http://www.reddit.com/r/MachineLearning/comments/2lmo0l/ama_...

Bill_Dimm··on Deep Neural Networks Are Easily Fooled
"The results are not specific to neural networks (similar techniques could be used to fool logistic regression)."

Do you have any references or examples where such behavior has been demonstrated on non-NN algorithms (like logistic regression)?

Bill_Dimm··on Top Martin Gardner Physics Stumpers
"...calculate how far apart they are one minute before they collide"

So, the answer is 500 miles (not 1000 miles).

Bill_Dimm··on Stop Changing UIs for No Good Reason
Yep, I still find myself switching to the old version so I can get terrain view. In the new version, once you type in an address the "terrain" link below the address box disappears. I can't imagine why that is supposed to be helpful. My typical use case is that I click a map link for a hike, enter my address to see how long the drive is, and then want to switch to terrain view to see terrain for the hike. Apparently, that is no longer possible.
Bill_Dimm··on How Drinking Bottled Water Became a Thing
It's been done (not $0.25, though): http://grrrltraveler.com/top-packing-tips/vapur-water-bottle...
Bill_Dimm··on What makes for a stable marriage? Part 2
It's a shame the "being the same age" item didn't separate the data for men being older from the data for women being older (for heterosexual marriages). With society being somewhat more accepting of older men marrying younger women rather than the other way around, it would be interesting to see if marriage success agreed with that or not.
Bill_Dimm··on Please remove mitsuhiko/*
I thought DMCA applied to copyright (hence the "C" in the name), not trademark. Am I wrong?
Bill_Dimm··on YogaGlo Patent Dropped
The title is wrong and misleading. It should be "YogaGlo Patent Dropped." They did not patent yoga; they patented a method of filming yoga classes.
Bill_Dimm··on Nearly Every USGS Topo Map For Free
No, I haven't. Thanks!
Bill_Dimm··on Nearly Every USGS Topo Map For Free
Perhaps, but they seem to be more up to date than the maps I bought from Garmin for my hiking GPS a few years ago. I bought the 2008 version of their 100k maps for the U.S., and found that Lake Nockamixon (1,450 acres) and Marsh Creek Reservoir (535 acres) were both shown as the tiny creeks that were present before they were dammed up to create the lakes in 1974 and 1973 respectively. Imagine that you are hiking along and encounter a lake that is a few miles long that is not supposed to be there. Those lakes are showing up just fine on the USGS site.

When I complained to Garmin, their response was that it was not an error and I should buy the (more expensive) 24k maps, which do show the lakes. They claimed the 100k maps don't have enough "detail" to show the lakes, yet they do show the tiny little creeks that were there 40 years ago, and they show other lakes that are only 22 acres. So, Garmin will happily sell you maps based on 40-year-old data and then charge you more to upgrade to newer data. By comparison, the USGS maps seem stunningly timely.

Bill_Dimm··on Best Papers vs. Top Cited Papers in Computer Science
Nitpicking, but why are they claiming to provide MAP (mean average precision) scores when their description and equation indicates that they are computing average precision, not MAP. According to the definition of MAP [1] that they link to, MAP is computed across multiple queries while average precision is computed for one [2]. Furthermore, they truncate their calculation to only consider the top 3 cited papers (i.e., they don't go all the way to 100% recall), so it's not even really the average precision.

[1] http://en.wikipedia.org/wiki/Information_retrieval#Mean_aver...

[2] http://en.wikipedia.org/wiki/Information_retrieval#Average_p...

Bill_Dimm··on From the startup who allegedly stole software and raised $2M with it
The problem I have with that argument is that the original post said there were multiple payments that were all reversed later. If the code was so bad, why wasn't the developer terminated much earlier? Doesn't multiple payments normally imply that multiple milestones were hit (i.e. deliverables accepted)?
Bill_Dimm··on It's not too late to ditch the ad-based business model and build a better web
So, AOL? When AOL merged with Time Warner they made the various Time Warner magazine sites subscriber-only unless you were on AOL.

Edit: Oh, and what makes you think the advertising wouldn't come back eventually? When cable TV came out one of it's big selling points was no ads. Then, someone realized: "Hey, we can make even more money if we sell ads on top of the subscription fee." Nobody is going to resist the temptation to stuff ads into every nook and cranny unless it causes them to lose more revenue from losing customers than they gain in ad revenue, which never seems to happen.

Bill_Dimm··on Restaurant getting bad reviews for slow service, hired a firm to investigate
Customers are spending nearly twice as long at the restaurant because they are busy with their phones, yet the restaurant says "the number of customers we serve on a daily basis is almost the same today as it was 10 years ago." How can they serve the same number of customers if each is taking twice as long? Were half of their tables empty 10 years ago?
Bill_Dimm··on 7 Reasons our app costs $99.99
The charts are interesting, but I wonder how meaningful they are. For example, does the sales volume at $9.99 reflect what long-term sales would be, or is the number artificially high because people saw the price drop and thought they should buy immediately in case the price went back up. Since the $19.99 price followed the $9.99 price, is that number artificially low because the $9.99 price already drained potential customers from the market? Of course, none of that argues against the conclusion that $99 gives them more revenue than $9.99.
Bill_Dimm··on Improved default settings for Linux machines
It sounds like you had a process running over night that needed enough memory to force X (or whatever) to swap. Without the swap, that memory-hungry process would presumably have run less optimally (if it got memory allocation failures and adjusted its behavior to utilize less memory) or it would have simply died. It's not so clear that having processes die because they can't get memory is better than having something else run slowly, in general.

Edit: Possible caveat: Maybe whatever was running overnight did a lot of disk I/O and the OS decided to cache it at the expense of moving idle processes to swap (not sure if Linux does that or not).

← PreviousPage 2 of 7Next →