Fig. 3
From: LightGBM: accelerated genomically designed crop breeding through ensemble learning
Comparison between LightGBM and rrBLUP. a Comparison of three variants of GB algorithms, namely LightGBM (LGB), XGBoost (XGB), and CatBoost (CB), with GB and rrBLUP in terms of the model fitting ability. b Comparison between the five methods in terms of the computing time and memory usage. c Improvement in the model precision of rrBLUP (rrBLUP+) and LightGBM (LightGBM+) via the addition of parental phenotypes as additional features. Left panel: 6210 F1s + 1221 Zheng58 F1s as training samples to predict 1221 Jing724 samples. Right panel: 6210 F1s + 1221 Jing724 F1s as training samples to predict 1221 Zheng58 F1s. d Use of two coding schemes (0, 1, 2 and 0–9) for converting genotypic characters to numeric features to compare the precision of LightGBM. Predictive framework: 6210 F1s + 1221 Jing724 F1s as training samples to predict 1221 Zheng58 F1s