1) In the demo version, I used the default data to get an statisfory solution as follows with R2 about 0.96xx in Train/Test (75:25) under the random seed (0)
Size Error CV_Error Function
15 0.285075 0.302654 (0.09430305720946518x-5.083683630458345)cos(-0.09963877022688811*x-(-1.565274357571482))
2) When I rerun in a search without cross-validation (aim to show how good the full set of data fit the solution) by reloading the solution above, which shows the solution with an R2 of a low value (something about 0.59xx). I think R2 should be something about 0.96 as close to the R2 value for both train and test data of the solution obtained in the split run.