How would you do it?

[Peter.Greulich]

Re: How would you do it?

...Just a cold front moving in changes everything---there is no way to predict the weather based on what happened today or even yesterday. ...

Mike, you got a point there. This may no longer be true, but years ago there was a study comparing weather forecasts between the U.S. Weather Bureau, the Farmer's Almanac and astrological forecasts. Yes, you guessed it right! The Farmer's Almanac came in first, the astrologer second, and the Weather Bureau last.

[MikeC]

Re: How would you do it?

LOL...even now I can predict for my local area better than the forecasts given generally as they tend to have to encompass a much larger region and give a composite not a specific forecast. The Farmer's Almanac would be better than trying to use the much smaller datamart Richard intends or is thinking of using....but website data can be obtained automatically and, as far as I know, the Almanac's data would have to be entered in manually.

[gww]

Re: How would you do it?

My gut feeling before I responded to the first post is that you might be better served using a statistical package like SAS, SPSS, or S. Why do you wish to use Alpha 5 and have you used any statistical packages before?

[MRichard]

Re: How would you do it?

My gut feeling before I responded to the first post is that you might be better served using a statistical package like SAS, SPSS, or S. Why do you wish to use Alpha 5 and have you used any statistical packages before?

Hi GWW, and thank you for mixing in this conversation!
You have a legitimate question.
However, the application covers a lot of stuff that is better off being dealt with in a database program and which would hardly be possible in a statistical package.
Now, you could export data to any statistical package and from there do the further math and analysis with that data to determine correlations, regression, extrapolation etc.
But as a programmer you want to "centralize" things (if possible of course) in 1 app to make it easier on your users in terms of maintenance, installation, instruction, navigation etc etc.
And it is not to far from the bed to think that a database program may cover data manipulations best..... :-)
I see your point though, and there ARE several (even open source) statistical packages that would do the job more easy, I believe that. However, I also think that Alpha could do the trick, as long as we tell it precisely how to do it.
And if I manage to, would the end result not be better having it all in just the one application, then knotting all kinds of applications to each other ?

[MRichard]

Re: How would you do it?

LOL...even now I can predict for my local area better than the forecasts given generally as they tend to have to encompass a much larger region and give a composite not a specific forecast. The Farmer's Almanac would be better than trying to use the much smaller datamart Richard intends or is thinking of using....but website data can be obtained automatically and, as far as I know, the Almanac's data would have to be entered in manually.

Hi Mike, it probably is true, that the forecast of the weather brings a better prediction of the weather then just your own historical very narrow based data. But what next? Regardless of the source of the data, being the Weather forecast service or your own dataset, how to relate the data in the tables to each other? How to predict the data in the result table outside the known values until today?

[G Gabriel]

Re: How would you do it?

Richard:
Your quest is understandable and reasonable, but let's just make sure we have some basics covered:
1-You cannot (nor anyone can) "Predict" the next scenario
2-You can make a "Best Guess" of the most likely scenario thus creating a cone of error or a Bell curve of probability
3-You cannot turn art into science. Most predictive models rely on "science" which is no science since built in its foundations are variables chosen with no scientific basis. To illustrate, let's say you wish to use a Moving Average, the question becomes:
a-Do you use a Simple MA? Or
b-An Exponential MA?
c-And if you do, what period do you apply?
d-And if you use an Exponential MA, what Smoothing Factor do you use?
The answer to any of these is purely empirical and experimental.

So, YOU have to decide what do you want to use, and then write a function for that. You can easily write one for SMA, EMA or anything else. It's all mathematics. The question really is up to you: what do you want to use? Then comes how to write a function for it.

As I suggested before, you might want to start with an SMA then apply SD to that. My personal observation has been, in a large data pool with a long track record like weather data, EMA is not superior to SMA nor has any more predictive value, but you could do it either way.

You most certainly can create statistical functions in alpha. As I said earlier, it's all a matter of mathematics. I started writing one a while back for Linear Regression but didn't finish it. But all is doable.

[MikeC]

Re: How would you do it?

Richard,

how to relate the data in the tables to each other?

I say you don't initially as many times there will not be any distinct correlation between them. Determining how each affects the visitor numbers is key. A simple line graph would give you a very good idea of how each should be weighted---the ones that show more of a matching increase/decrease are the ones that would have more weight and influence the end result more....this is to be done simply to determine the amount of influence each parameter has as once found, the rest is simple multiplication and addition.

I am thinking that even visually doing this, with enough data, that it would come very close to what you want and even could be just as valid as whatever you come up with using other mathematical methods.

As G states, you have to determine which way to go and then use it--personally I would try simple and fast and then go more complex if not predicting well enough.

I will be curious as to what you finally end up using.....and then compare that with the graphing method.

[robert]

Re: How would you do it?

I think the best way to do the prediction is to "look out the window".

Seriously, the relationships between the data in Tables A-D to produce an outcome has had to be done before. I don't think it's a question of whether you can use Alpha to produce Table E. You simply have to give Alpha the relationship which is on you to determine. Analysis of Variance can be used to give you any result you want.

BTW, why do you need four tables when you can input all the data into one?