odbc_connect() is giving me no shortage of other issues as well.

Yeah I was defending PHP. I use it as my primary scripting language. Doesn't mean there are things I don't hate about it.

I have a MBP and recently it has been getting ridic hot near where the power adapter plugs in when charging. Today it seems to have totally stopped charging when plugged in (no light comes on when power cord is attached).

Does this sound more like a battery issue or power adapter, or some sort of other hardware problem? Any ideas?

neko,

apple store -> appointment -> genius bar


wooster,

as i understand it, the principal and huge problem with fb advertising is that you have zero engagement with the userbase. they're not on fb to buy things or learn about new products.

is it a programming question or a design of experiments question?

If what you want to know are how relevant the variables are to the end product, you'd generally want to do a regression analysis or supervised learning with a neural net since you know the outputs.

I had no idea you needed to reserve an appointment at the apple store (never been in one before ) but I made an appointment for tomorrow. Thanks.

Was able to get it charging a bit by holding the cord carefully and not breathing so it's looking like it's maybe just a flaky connection in the cord.

It sounds like you are just trying to some data fitting?

For example you have a bunch of data points (x,y) that you know come from a function that looks like y = a*x3+b*x +c*e^-x and you want to figure out what the a,b,c coefficients are?

This is pretty straightforward if so.

rh2,

i guess my answer is that if you want to talk about the mathematical/statistical aspects of the problem, put it in smp. if you want to talk about the mechanical/algorithmic/programming aspects, start a thread here.

Is there data-based proof of this assertion? Some of the most popular pages on facebook are corporate and celebrity (business?) pages.

I would agree to a certain extent that facebook is best used for keeping clients and improving client satisfaction and feedback, however, that doesn't mean that it can't be used to gain new clients via advertising and what-have-you.

The main gist is the same as a company that slaps up a website and doesn't consider usability or company image, fully expecting the drag-and-drop site builder to save the company. Obviously, this doesn't work out, and this correlates perfectly to facebook: all the advertising in the world means nothing w/o conversion.

The problem is that, in my opinion, people marketing, including professionals, don't have any notion of the goals of their plan, and without some prescribed measurable metric, they are going to be disappointed because they automatically assume best-case when they should be considering (almost) worst-case.

The point is that there ~is~ data available that suggests running ads on facebook is highly profitable, but the caveat is that getting visitors amounts to nothing special in itself. It really breaks down to "how much can one afford to spend on gaining one client" and "what do I do once the client clicks my page?"

Here's one source: http://www.hindustantimes.com/techno...e1-856594.aspx

Sorry, 17% plus some x number is clicking on or "sometimes" clicking on ads? That's really good, actually. Even in the portals my company advertises on, we don't see anything close to numbers that correlate to this, and this is on portals that people go to for no other reason but to shop and find new products.

However, the main gist is that it all breaks down to how much you can afford to gain one customer, and that is the guiding principle to every marketing decision you make.

An example of one campaign I was working on: Best-case, try starting with one out of every 1,000 impressions equal one click-through, and out of those, 1 out of 10 (law of "not trying at all" in sales") will buy one product, thus consider 1 out of 10,000 impressions will convert to a customer. Multiply that number by impressions. Suppose it's $0.05 / impression, and your total is $500 for one customer.

The above is, IMO, best-case, and not truly an attainable goal for someone's first ad campaign, thus, you may consider your best case in this to something like 1 out of 30,000 impressions, then figure out if you can afford to spend $1,500 on each customer. If you're a cruise line, absolutely. If your selling an e-book, absolutely not.

I'm pretty sure facebook has all the metrics available for impressions and click-through. I would probably cut the number they give in half and then figure out your client-conversion rate, but consider it to be worse than 1 out of 10 as well. So, suppose, facebook tells you that click-through is generally 1/10% for each campaign. I would consider something closer to 1/20% and then client conversion to 1 out of 20 and plug in the numbers from there.

dave,

i upvote your answer. i know basically nothing about marketing and my remark was off-the-cuff based on a conversation at work with other people who know nothing about marketing. i should have qualified my answer better.

Next time I post something like, this, please tell me to get my head out of Scheme's ass and stop building my own primitives. kthanks.

****
Continuing with the SQL questions.

I just installed the PostgreSQL (I know, I know) database dependencies for this project. I guess I'll try out sqlkorma, and ORM for Clojure. Seems pretty nifty and looks like it could be pretty fun to use.

Is there any objections to ORMs in general?

if you're working with tons of records, they can slow things down since you have to build all the objects, but typically they are a big positive

Would you be happy to replace the requirement that 53 and 7 be the median with 53 and 7 being the expected value? Because that's a really easy problem. It can be done entirely with one type of scoring--e.g. only touchdowns or only field goals.

Let's pick touchdowns. Say p1 is the chance that team1 scores a touchdown in any given possession and p2 is the chance that team2 does. Say there are n possessions. Then we just have 7n*(p1+p2) = 53 and 7n*(p1-p2) = 7. Adding the two together, you get 14n*p1 = 60. So, p1 = 30/(7*n). p2 = 23/(7*n).

If you want to solve the exact problem that you gave, then you do indeed have to add in field goals, and it takes more work than I'm willing to do right now, but it's still just an algebra exercise. If you post it in the probability forum, you'll have an answer very quickly.


BTW, I really hope you're not planning to use this to bet on sports. It will not yield anything fruitful...

Facebook Ads are certainly profitable (if you do it right - which means a ton of testing/tracking and analytics), otherwise there'd be nothing but "branding" on there from supercorps with more money than sense. As is, it's full of independents running campaigns (mostly to dating sites as far as my male/single FB profile can see).

Basic rule of internet ads - if you see something running frequently and for more than a little while, you can be pretty sure the person behind it is making money, or is a ******.

For those interested in learning a lot about reasonable sized / profitable internet ad campaigns, this 2p2 BFI thread is amazing: http://forumserver.twoplustwo.com/30...-amaa-1208283/

Or if it's classification support vector machines. But yeah what you're looking for is basially "learning" in an AI sense...once again good chapter on that in "Artificial Intelligence, a Modern Approach"

Re: Facebook...I think the problem is that there are basically a bunch of sockpupped clickthrough agents. Can't provide any data to back up that claim but I'd say the quality of clickthroughs is pretty low.

FWIW, the answer with expected value is fairly close to having a median where you want it.


7n*(p1-p2) = 7 is the formula setting the difference between their scores properly. 7n*p1 is team 1's expected score. 7n*p2 is team 2's expected score. so 7n*p1 - 7n *p2 is the difference.

Because bookies aren't idiots and you can't beat them by extremely simple arbitrage like that. Bookies took a beating a while ago when people started throwing some basic math at the question. Now, the lines are set extremely carefully. There are still people beating the bookies, and some with relatively simple strategies, but nothing like this could possibly get.

Plus, it's a terrible model. You ignored turnovers, for example. Modern bookies certainly have a much better model than that.

If you do plan on doing this, at the very least, test it on a large dataset before putting any money into it. If you don't have a large dataset, do something else with your time.

You have a very naive view of how sportsbooks work. This statement clearly shows you don't understand how pricing is set. I suggest you think about it a little more. If Pinnacle handles millions of dollars in volume per game, do you really think the liquidity there won't beat it into a fair price?

There are numerous economists who spend most of their waking hours building better models to predict the outcome of sporting events in various ways.

Bookies across the world are paranoid about someone beating their lines and work feverishly to prevent it. They do a ton of work in modeling the games themselves, and they also move the lines when they notice that sophisticated/large volume bettors are betting asymmetrically.

They also take a huge vig, which is ridiculously hard to beat.


Again, I'm not exactly saying not to try this system. I'm saying to try it by testing it on a large dataset. If it turns out that it works, you can say I told you so or just keep it to yourself and make some money. I guess you might regret the small amount of time that you'll have wasted testing. But, you're not right, so testing will scratch your itch without costing you money and will embarrass you less.


Yes. You should look into Python(x,y), which comes with iPython, a ton of packages such as the three you mentioned., and a great Python IDE called Spyder.

It's fine to learn. But you are learning something so rudimentary that won't yield "profit." That is not the point of doing introductory exercises.

I'm currently a security consultant for poker sites. I spend pretty much all of my time analyzing large datasets with numpy/scipy/python.

I haven't tried everything, but from what I can tell, this seems to be by far the best environment for my work, which tends to be very experimental and incorporate lots of different ideas, many of them totally outside of statistics.

For stuff that is purely statistical, a more specialized language such as R (which I've used and like) would be better. It's slightly simpler to do basic statistical things in R because many standard statistical functions are built-in, so you don't have to find the right package and import it. R also has a lot of really advanced stuff; I've had to reinvent the wheel a bit in Python because there simply aren't publicly available packages for some slightly obscure statistical concepts, but I always Google it first and see that R has a package and want to kill myself. I've also noticed in my Googling that Matlab seems to have a lot, but I have no experience with Matlab.

If you're working with tons of data (maybe on the order of 10M datapoints or so, but it really depends on the application and the algorithm--I've used Python on 100M datapoints and I've had to use something other than Python on 10k datapoints), Python gets to be pretty ridiculously slow unless you're really good about worrying about speed. Even scipy and numpy, which are supposed to be designed for this, have some really frustrating bottlenecks in some of their algorithms. In that case, you might want to try using C++. But, that really sucks for doing research/experimental stuff because you have to keep recompiling and reimporting data.

People who have good sportsbetting ideas typically don't share them. You might try looking up some old ideas that used to work to get you started. For example, it used to be true that bookies didn't account for the distance that the away team had to travel before a game. It turns out that sitting on a plane for 6 hours the day before a game lowers your chances of winning a bit. You might try going back through old data and proving that or thinking about other similar things that might be true. You might consider the fact that bookies now include tiny facts like that in their models at the margins and realize just how incredibly hard it is to beat them.

Edit: I'm not a good guy to teach you about programming or statistics. I'm a good guy to teach you about how to catch cheaters on poker sites, but I'd have to kill you afterwards unfortunately.

The direction you should go in is learning how to do the basic stuff (scraping data, combinatorial analysis, regression, monte carlo simulations). That is its own reward. You aren't going to get rich on learning this stuff. It's very unlikely you'll develop a model that beats sportsbetting for any money. (And if you want to beat SBs, just chase steam. Much easier.)

Sure it's possible. Just that it's very hard to do and you have a long way to go before you can consider it.

django or pyramid or pylons

I wish the Programming mods would give this guy his own containment thread like the SB mod did.