Skip to main content

Table 3 Details of parameter settings for GNN models

From: Automatic disease prediction from human gut metagenomic data using boosting GraphSAGE

Parameters

Bagging & boosting GraphSAGE

Bagging & boosting GCN

Number of convolution layers

2

2

Number of hidden units in each convolution layer

32

32

Sample number for layer 1

10

–

Sample number for layer 2

5

–

Activation function - convolution layer

ReLU

ReLU

Activation function - output layer

Softmax

Softmax

Epochs

50

50

Optimizer

Adam

Adam

Learning rate

5.00E−03

5.00E−03

Loss function

Categorical cross entropy

Negative log likelihood