Hi all,

I'm hoping someone might have an answer for my query. I've got a NN that I use for doing sports predictions, and in my latest tinkerings, I've encountered a problem where my results seem correct, but the values are of a much smaller scale than I would have predicted.

I'm using the number of consecutive wins/losses for the input, and a 1/0 to determine if the team won or lost that particular game after x wins/losses. These are the results that I have been getting. The top graph is the result that my NN has been returning, the bottom is the statistical graph of the same results. The yellow (Punter, the name of my NN) is the scaled up version of the top graph. As you can see this result seems to fit the situation, but not to the same scale.

I'm not sure, any ideas? As I said, I can quite easily scale this up, but the inputs for a win/loss are 1/0, so I would have expected something closer to those figures for the output, if it's actually learning anything.

If anyone's interested, the current system is up on my website: http://punter.freespaces.com, but not much techy info unfortunatly.

Cheers,
Hoopa