BMC Genomics

Table 2 The comparison of the performance when a constant learning rate and a decreasing learning rate are applied

From: A maximum likelihood algorithm for reconstructing 3D structures of human chromosomes from chromosomal contact data

	Constant Learning Rate		Decreasing Learning Rate
Input Parameters	Running Time	Accuracy (Average dSCC)	Running Time	Accuracy (Average dSCC)
CHR = 1-22, NUM_STR = 1, ALPHA = constant	4 min	0.821	13 s	0.8493
CHR = 1-22, NUM_STR = 1, ALPHA = [0.1, 2]	1 h, 30 min	0.8456	3 min	0.8536
CHR = 1-22, NUM_STR = 5, ALPHA = [0.1, 2]	7 h	0.8546	20 min	0.8546
CHR = 1, NUM_STR = 1, ALPHA = constant	37 s	0.7556	2 s	0.8088
CHR = 1, NUM_STR = 5, ALPHA = [0.1, 2]	1 h	0.7841	3 min	0.8088
CHR = 21, NUM_STR = 1, ALPHA = constant	0.7 s	0.8969	0.2 s	0.8995
CHR = 21, NUM_STR = 5, ALPHA = [0.1, 2]	36 s	0.9018	2 s	0.9018
CHR = 21, NUM_STR = 30, ALPHA = [0.1, 2]	4 min	0.9018	12 s	0.9018
CHR = 21, NUM_STR = 50, ALPHA = [0.1, 2]	6 min	0.9021	18 s	0.9018
CHR = 21, NUM_STR = 100, ALPHA = [0.1, 2]	12 min	0.9020	37 s	0.9020
CHR = 21, NUM_STR = 200, ALPHA = [0.1, 2]	24 min	0.9022	83 s	0.9020
CHR = 21, NUM_STR = 500, ALPHA = [0.1, 2]	1 h	0.9022	3 min	0.9021

The comparison of the computing time and the average dSCC value obtained by using a constant or a decreasing learning rate for different input parameters for the chromosome 1 – 22 of the GM06990 cell line. We used the constant learning rate 0.0001, and we defined the initial_λ = 0.01 for the decreasing learning rate. CHR represents the chromosome number, and NUM_STR represents the number of ensemble structures generated per conversion factor(α), ALPHA represents the conversion factor. The decreasing learning rate achieved a better computing speed in all the cases

Back to article page

ISSN: 1471-2164

Contact us

Submission enquiries: bmcgenomics@biomedcentral.com
General enquiries: ORSupport@springernature.com