Skip to main content

Table 1 Optimal set of hyperparameters for the LMCL function

From: FoldHSphere: deep hyperspherical embeddings for protein fold recognition

Model

(a) LMCL

(b) Thomson LMCL

Scale

Margin

Iter THL-sum

Scale

Margin

CNN-GRU

30

0.25

1130

30

0.25

CNN-BGRU

30

0.55

1172

30

0.45

ResCNN-GRU

30

0.50

1181

30

0.55

ResCNN-BGRU

30

0.60

1020

30

0.60

  1. The scale and margin hyperparameters are provided for each neural network model and two approaches: (a) training the last classification layer end-to-end, (b) using the fixed prototype matrix by minimizing the Thomson loss THL–sum. We also include here the optimal iteration from the Thomson algorithm
Back to article page

BMC Bioinformatics

ISSN: 1471-2105

Contact us