Hi,
I posted this in the other forum too. I'm working on building a generalized model for predicting scores for various benchmarks based on a number of attributes. I've so far collected about 3600 scores from different reviews and right now am working on verifying that the model seems to be working as it should. It's actually predicting *very* well when doing a 66% split or 10 fold cross validation.
Right now I'm getting a mean absolute error of about 8%, and a confidence of 0.996 (this is using M5 rules in Weka). Once I've verified that everything is working as it should (I need to make sure that the model isn't overfitting the data or anything) then I'm going to work on the paper, and eventually I'm hoping to get some sort of website built to play with the model online.
In laymens terms, I'm hoping to make a website where you enter in a cpu/videocard/memory configuration, and a benchmark with a certain map/resolution/AA/Aniso/etc, and you'll get a predicted score within a certain degree of accuracy.
Thanks,
Nite_Hawk
I posted this in the other forum too. I'm working on building a generalized model for predicting scores for various benchmarks based on a number of attributes. I've so far collected about 3600 scores from different reviews and right now am working on verifying that the model seems to be working as it should. It's actually predicting *very* well when doing a 66% split or 10 fold cross validation.
Right now I'm getting a mean absolute error of about 8%, and a confidence of 0.996 (this is using M5 rules in Weka). Once I've verified that everything is working as it should (I need to make sure that the model isn't overfitting the data or anything) then I'm going to work on the paper, and eventually I'm hoping to get some sort of website built to play with the model online.
In laymens terms, I'm hoping to make a website where you enter in a cpu/videocard/memory configuration, and a benchmark with a certain map/resolution/AA/Aniso/etc, and you'll get a predicted score within a certain degree of accuracy.
Thanks,
Nite_Hawk