Fig. 4
From: LightGBM: accelerated genomically designed crop breeding through ensemble learning
Sampling rate of training data determines baseline precision. a The number of 621 testing samples remained fixed, and nine sets of training samples with different numbers were randomly selected from the rest 5589 samples. Precisions of rrBLUP and LightGBM were then compared at the nine ratios of training and testing samples. b The number of 621 training samples remained fixed, and nine sets of testing samples with different numbers were randomly selected from the rest 5589 samples. c The ratio of 9:1 of training-to-testing samples remained fixed, and different numbers of samples were randomly selected from the total 6210 samples. Precisions of rrBLUP and LightGBM were then compared at the nine sets of training and testing samples. d The ratio of 1:9 of training-to-testing samples remain fixed, and different numbers of samples were randomly selected from the total 6210 samples