Shannon's Journal Discussion

[Shannon]

A long time ago, I started trying to develop a computerized predictive model for helping me figure things out faster. Instead, for example, of trying 30 million variations of something, it helps me narrow down my efforts by showing me what will probably happen if I use each option.

Over time it's gotten more and more complex and refined. Now it consists of three models that are all using slightly different algorithms, which all have merit and are accurate. But when two or more of them agree within about 5%, or three of them all agree within about 10%, I know they are around 85% likely to be correct.

When I run various questions that are related through my models, they show me with a very high degree of accuracy what is most likely, because the effect is that one can't be right of the other three disagree. It ends up being a little bit like how the oil companies used to map out oil deposits by drilling a series of cores and then building a computer model from them.

The original model was based loosely on the idea of weather modeling. Over time, I have refined it and it is more and more becoming something that has very little relation to anything else I know about. Basically takes into account a lot of data and then manipulates that data in special ways to arrive at an answer which it outputs.

I'm going to have to rewrite the models again soon, have been lazy about that. Current models need to be refined for more speed and ease of use.

But when I say models, that's what I am talking about.

Do you use some known algorithms as foundation? Or maybe neural networks? And where do you get data for things like meeting a certain person? Just curious.

I use algorithms I developed myself. Beyond that, I'm keeping it a secret.