. 2021 May 17;22:252. doi: 10.1186/s12859-021-04163-y

Table 3.

Range of training set sizes that was used to fit the power law expression in Eq. (1) and the goodness-of-fit measure, $M A E_{fit}$

Dataset	$m_{k = 1}$	$m_{K} = \| T \|$	$M A E_{fit} (\times 10^{- 5})$
Dataset	$m_{k = 1}$	$m_{K} = \| T \|$	hGBDT	sNN	mNN
GDSC1	20,000	115,863	2.0	8.4	3.3
GDSC2	20,000	78,423	0.8	2.1	2.6
CTRP	50,000	203,650	0.4	1.0	1.7
NCI-60	100,000	675,000	0.8	2.1	0.9

The $R_{fit}^{2}$ for all the listed experiments is higher than 0.99