Skip to main content

Table 5 Hyperparameters chosen for training of the neural networks are shown here

From: Classification of substances by health hazard using deep neural networks and molecular electron densities

Hyper parameter

ECHA dataset

CompFood dataset

EPOCHS

38

14

Learning rate

0.000844

0.0002409

Batch size

20

12

Rate decay

0.1 every 35 epochs

–

Weight decay

2.57e−7

0.000186

Feature size

28

4

Filter size

4

4

Weight class 0

0.65127

1.24926

Weight class 1

1.34672

1.1977

Final layer neurons

16

32

Back to article page

Journal of Cheminformatics

ISSN: 1758-2946

Contact us