Interface for printing the network to file Interface for printing a summary of the network Interface for reading the network from a file Interface for reading network settings from a file Interface for reading optimiser settings from a file Interface for building network from ONNX nodes and initialisers Interface for adding a layer to the network Interface for resetting the network Interface for compiling the network Interface for setting batch size Interface for setting network metrics Interface for setting network loss method Interface for setting network accuracy method Interface for resetting state of recurrent layers Interface for saving input to network Interface for saving output to network Interface for training the network Interface for testing the network Interface for returning predicted results from supplied inputs using the trained network Interface for returning predicted results from supplied inputs using the trained network (graph input) Interface for updating the learnable parameters of the network based on gradients Interface for generating vertex order Interface for depth first search Interface for calculating root vertices Interface for calculating output vertices Interface for reducing two networks down to one (i.e. add two networks - parallel) Interface for copying a network Interface for getting number of learnable parameters in the network Interface for getting learnable parameters Interface for setting learnable parameters Interface for getting gradients of learnable parameters Interface for setting learnable parameter gradients Interface for resetting learnable parameter gradients Interface for forward pass
| Type | Intent | Optional | Attributes | Name | ||
|---|---|---|---|---|---|---|
| class(network_type), | intent(inout) | :: | this |
Instance of the network |
||
| integer, | intent(in) | :: | epoch |
Current epoch number (1-based) |
||
| real(kind=real32), | intent(in) | :: | loss |
Current loss value |
||
| real(kind=real32), | intent(in) | :: | accuracy |
Current accuracy value |
Hook called after each training epoch. The default implementation is a no-op; override in a derived type to add custom per-epoch behaviour (e.g. W&B metric logging).