Cool!
But how is it possible that my rule worked with “120% of the 153 colours everyone has seen”?
Also, it seems I can only use AND once?

Tessa: I may have made a mistake (nothing like a bunch of people to expose problems carefully planned testing didn’t). It should be fine now

jg: We’re concerned with not just matching what is blue, but also correctly matching what people say is not blue. Your rule has to work both ways, and so is judged both by the system.

How do the percentages work? I have a rule that says a “60% not blue” is not blue, but I am not getting credit for it. Just curious how the percentages factor into it.

Awesome – I’ll be turning my students loose on this one.

And had this been next year I would be requiring in both my 8th grade intro to programming and 9-12 Foundations of Computer Science classes. Very, very cool idea.

I think what is most frustrating to me is that I work to get a high percentage on my 30 colors but for some reason it does not translate well to all the colors.

I have got multiple 90% and higher but the highest I have seen in the total results was 87%. (I can’t remember which was which but I think that was even my first attempt, which showed a local percentage of 84%.) It seems my equations suffer from being tailored to the few colors I have seen rather than being a general expression of ‘what is blue.’

I’m having way too much fun setting up my logical statement/inequality.

A really great related article is XKCD’s color survey from a few years back, where they asked people to name colors. Warning: there are a fair number of not-safe-for-math-class words, mainly in quotes that shows what happens when you ask anonymous people to fill out online forms.
http://blog.xkcd.com/2010/05/03/color-survey-results/

A couple of minor thoughts — this is, in general, pretty great.

– Why is everything on the leaderboard rounded to the nearest percent? With the data you have, you could give a little more accuracy (two decimal places) and spread out those “ties”.
– The convention on those tied places is to mark the next place according to the number of ties: 1, 2, 2, 2, 5, 5, 5, 5, 5, 10, 10, 10… rather than 1, 2, 2, 2, 3, 3, 3, 3, 3, 4, 4, 4… Otherwise you’ll have everyone above average… which I suppose is good for your VAM.
– Math question: is there a way to combine a statement like (b > 200 and g > 200) into a single inequality? b * g > 40000 doesn’t cut it, and neither does (b-200)(g-200) > 0.

Hi,
Doesn’t work using firefox 21.0, can’t drag the numbers, just so you know that’s all.
Steve.

Hi Matthew. Nice catch! It should be working fine now.

I input this rule, bg, and the program still gives me a high (87%) rating. How/why is this possible?

Thanks, Scott

I can’t seem to use parentheses in the formula. What is the order of precedence for And/or? Or is that part of what i need to get the kids to figure out, hmmm. Think I will let the kids puzzle over this. Thanks I have some kids really engaged.

Great problem.

and is the formula is trying to work with all of the ones that say 100-50% blue, and none of the ones that are 50-100% not blue? or every single one?

@Dan, thought I responded earlier. The kids will be together for the homeschoolpizzaparty (if they win) we meet in person usually and occasionally online. Online never works as well.

Also found a bug, I entered “r=1” and got a 57% match!!!

If you send me an email with the code for your parser I can try and help (time permitting)

Clarificationon the Parentheses issue:

(r+g) > 40 <== Works

r < 10 and (b 200) <== Does Not evaluate

@Steve: r=1 will return false for practically every colour…or not blue. This means that it will work for all the non-blue colours. Statically, this will probably get you a decent percentage across the entire dataset, but not as good as one that returns true (or blue) for blue colours as well.

Re. order of precedence: the ‘computer’ runs on logical AND before logical OR. Parentheses currently won’t work. I’m going to try and put some time in on this today. If it is around a single clause it evaluate, but won’t actually change the result.

I can’t seem to reproduce any of those problems. Is the blog stripping some characters here? (b 240) wouldn’t evaluate in any context. Can you drop me an email or something and I’ll take a closer look? dave@davemajor.net

@Jim: Sorry, I think you may have caught bluemalator at one of it’s busy recalculating periods and iOS timed-out. I’ve made some changes to hopefully reduce that risk. Hope it didn’t cost you a top score!

This is great. Such a nice idea.
I am far from a school kid, but I got into this. The reason for this is that this is a classical Machine Learning problem. As I completed Caltech’s online course “Learning From Data” a couple of months ago, this is the perfect exercise to brush up my knowledge and put it to the test!
Of course you know that this is Machine Learning, you are referencing the Netflix Prize ;)

It is great reading the comments and seeing people discover basic ideas of machine learning, such as a solution working well in the small sample size but not working well in the (hidden) overall data.

Initially I was a bit puzzled by the percentage of blue you report, since I thought we had o match this too. Then I realised that we only care about the binary decision. Maybe you can make this more clear (?)

Now I do not want to highjack your effort and its focus on school students, but would it be possible to release more samples of your set. You see, with only 30 samples it’s hard to do any fancy machine learning (although a quadratic plane does not perform too bad). If we could have 200 or 300 samples I bet we could do much better.
I could try using the data from xkcd’s survey, that would be interesting :)

@Dan – well if it makes you feel better two of my students beat me!!! It wasn’t until I asked them to share answers, that I was able to improve my score and beat them. I expect they will return the favor. No worries I have no intention of telling them what I am learning. Just trying to better understand the problem space so I can ask them better questions and try and guide them.

what is the formula?

So I see a “issue” with this test. Enter 1=0 (one of my students came up with this test). And you get 80%.

Why? Look at http://doyouknowblue.com/blueis

And count the # of blue in the test vs # of not blue.

@Thanassis; Good point and poor choice of words. I would use “1=0 gives 80%” to ask kids why? Then ask is it fair or a good judge of the “blueness” formula to have so many of the colors “not blue”? How would you judge the formula for blueness?

Thanks,
Steve Thomas (aka Mr. Steve)

@Steve: I see what you mean. This is an interesting point. The skewed frequency of classes is a well-known issue in machine learning. Generally you would have problems is for example only 1% of your data points were of one class and 99% were of the other class. There ways to account for this, but with 20% I think that the problem is not pronounced.

@Dave. Thanks. BTW I have a technical problem now, and maybe you can help me. I tried to submit a few rules earlier today, and the webpage seemed to freeze. I tried another rule few minutes ago, and again it took quite long, but it finally showed me the standings with my new rule in them. It also showed me the results from the earlier entries that I thought they never came through. There is one that did pretty well (3rd overall) but I am not sure which one was it :) Can you please send me the name of the teacher for my first 3-4 rules (just look for the name boulis in the standings)

When the site opened again after the first contest I decided to play a bit more with it. I saw that the original points/colours were kept. So I entered my best rule (a cubic plane) and it scored slightly better than before. Then I thought I’d try something major. You have a page (/blueis) that you give all the colors in the database in little boxes, separated in two regions blue and not blue. I saved these regions as images and then wrote a program to parse them and get r,g,b, values for all the different colours. So now I have the entire database (4090 points at the time). Now I could run the algorithm on the entire database and see what are the best results I could get. This is not really machine learning, it is more like fitting :)
Curiously enough the cubic regression on the 4090 points works slightly worse than the cubic regression on the 140 points I initially had!
Then I noticed some things that made me scratch my head even more.
I tried a 4th degree plane to separate the blue from non blue, as I was expecting it would do better than cubic with 4090 points. It did. With my calculations it was giving me an accuracy of 93.4%. But when I applied the rule with your website I got 80%. I assumed 80% is what you get if your rule compute always false. So maybe my rule was not parsed/calculated correctly (it is a long rule afterall). I checked this assumption by entering an always-false rule (r=300) and noticed that although it was close, it was not identical.
Moreover I started noticing that accuracy scores were changing in the standings *without* the number of colours changing. The number of responses are changing, but I assume, that the colours in the database are already fixed, i.e, when a color enters the database, it does not appear in the test and it does not change its blueness value.
Dave, if you have thoughts on the last points I’d love to hear them.

Steve, I think yes, it is by design. At least in terms of blue, you get almost the same colours each time. It makes sense to give 50:50 samples for the few colours the page shows. When you pick colours randomly then the ratio is 20:80 (that’s seems to be the case with the xkcd study too)

@Dave I believe there might be inaccuracies with the way you calculate the rules. Possibly due to rounding errors as I wrote above. I tried more tests with the database of 4304 colours. Using either 140 points or the full set of points (4304) I was calculating separating surfaces of various degrees (0th to 6th), and then testing their accuracy (i.e. how well they separated the blues from the non blues) on all 4304 points. The results I got should have matched the results from the standings page. But they do not. I only entered a few rules in the website, but all except for one (140 points, 3rd degree) perform worse compared to my expected results. The greater the degree, the worse they perform. The one exception performed 93.17% instead of 92.77% that I calculated.
In my rules I enter numbers with up to 14 decimal places of accuracy. Maybe your parsing truncates them to less accurate representations (?) I tried producing rules where the r g b values are divided by 100 first (so that the terms of a rule are larger and do not have to employ the e notation, e.g. 1.092344e-06). The results in the standings page are even worse (where of course my own calculations on accuracy did not changed). Anyway just reporting, probably it is not important for the rules school kids are playing with.