Table 1 WHEAT1 data set. Average Pearson’s correlations between observed and predictive values (and their standard deviation in parentheses) for seven methods for a single-environment model for 50 random partitions with 70% of the lines in the training set and 30% of the lines in the testing set. Methods GB, GK, and AK are the GBLUP, Gaussian Kernel, and Arc-Cosine Kernel, respectively. Methods Embedded Image - Embedded Image correspond to the Arc-Cosine kernel model with 1-4 levels (layers). The best predictive model for each environment (E1-E4) is in boldface
EnvironmentGBEmbedded ImageEmbedded ImageEmbedded ImageEmbedded ImageAKLevelGK
Single-environment model
E10.490 (0.04)0.520 (0.04)0.536 (0.04)0.544 (0.04)0.551 (0.04)0.561+ (0.04)110.561 (0.04)
E20.469 (0.05)0.474 (0.05)0.476 (0.05)0.477 (0.05)0.478 (0.05)0.478 (0.05)30.477 (0.05)
E30.378 (0.06)0.390 (0.05)0.400 (0.05)0.401 (0.05)0.409 (0.05)0.419+ (0.04)130.416 (0.05)
E40.450 (0.05)0.470 (0.05)0.482 (0.05)0.491 (0.04)0.491 (0.04)0.508+ (0.05)110.506 (0.05)
  • + Significant at the 0.05 probability level of the t-test for the hypothesis that the average of the correlation of kernel AK is superior to the mean of the correlation of kernel GB.