I'm currently doing a project on selecting the best learning algorithm to predict automobile prices from data on classified advertisements.I tried out couple of algorithms from scikit-learn and got some good R2 values.Then I tried neural networks from pybrain and neurolab and I wasn't able to get the R2 value above 0,even got negative values most of the time (I calculated the R2 using the metrics module in scikit).So got a couple of questions,hope you guys would be able to help
*is R2 a good criterion when trying to find the prediction accuracy of neural networks or non-linear models?
*What would be a good method/process to compare different learning algorithms and find the model with the best predictive ability?
[link][comment]