Optimizer Module

The optimizer class orchestrates the complete training loop for one model across all k-folds. It delegates data loading to dataloader, metric capture to metrics, and loss computation to the per-feature lossfx objects embedded in model_template.

Class: `optimizer`

Header: <generators/optimizer.h>

Inheritance: tools, notification

Public Fields

Field	Type	Description
`m_settings`	`settings_t`	Framework training settings (epochs, kfolds, batch-size, threads, …).
`kfold_sessions`	`std::map<int, model_template*>`	Maps fold index → cloned model instance.
`reports`	`std::map<std::string, model_report*>`	Maps model-name → training report struct.
`metric`	`metrics*`	Pointer to the metrics engine (not owned).
`loader`	`dataloader*`	Pointer to the data loader (not owned).

Public Methods

Signature	Description
`void import_dataloader(dataloader* dl)`	Sets the `loader` pointer and links `m_settings`.
`void import_model_sessions(std::tuple<model_template, optimizer_params_t>* models)`	Clones the model for each k-fold and initialises optimisers and schedulers from models.
`void training_loop(int k, int epoch)`	Runs one full training epoch for fold k.
`void validation_loop(int k, int epoch)`	Runs one full validation epoch for fold k.
`void evaluation_loop(int k, int epoch)`	Runs one full evaluation (test-set) pass for fold k.
`void launch_model(int k)`	Executes all epochs (train + validate + evaluate) for fold k.