Table 1 Performance measurement ratios. For each optimal set of hyperparameters in each subsampling, we reported the training time (in minutes and seconds), validation Accuracy (%), and both the overall test accuracy and the per-variant test accuracy (%).

From: Identification of mathematical patterns in genomic spectrograms linked to variant classification in complete SARS-CoV-2 sequences

Variant subsampling 01

Experimental hyperparameters

Training time

Validation accuracy

Test accuracy

Learning rate: 0.0001

Batch size: 52

Epochs: 1

1190 min 10 s

99.42%

Pre-VOC: 97.73%

Alpha: 99.79%

Delta: 99.88%

Omicron: 99.74%

Global: 99.40%

Variant subsampling 02

Experimental hyperparameters

Training time

Validation accuracy

Test accuracy

Learning rate: 0.0001

Batch size: 52

Epochs: 1

1626 min 16 s

99.52%

Pre-VOC: 98.12%

Alpha: 99.86%

Delta: 99.73%

Omicron: 99.85%

Global: 99.48%