Skip to main content

Table 4 Hyperparameters and training time of CNN models

From: Convolutional neural network models for cancer type prediction based on gene expression

 

Training

Testing

 

DL modela

Number of parameters

Loss

Accuracy

Loss

Accuracyb

Timec (seconds)

1D-CNN

211,489

0.01

0.9971

0.1769

0.9567

80.3

2D-Vanilla-CNN

1,420,737

0.007

0.9981

0.1778

0.9557

94

2D-Hybrid-CNN

362,177

0.0149

0.996

0.1586

0.9582

80.8

2D-3Layer-CNN

26,211,233

0.5149

0.9654

0.6875

0.9184

214.6

2D-3Layer-CNN (with patience = 10)

 

0.1976

0.9869

0.3914

0.9419

379.17

  1. aEarly stopping is used for all models (all with patience = 4, except for the last model)
  2. bResults of 5-fold cross-validations
  3. cAll models were trained using a Linux server with Xeon 8176 CPU @2.1GHz, with 4 × 28 cores
Back to article page

BMC Medical Genomics

ISSN: 1755-8794

Contact us