Task Runners¶

Task runners are Facade functions dispatched by the CLI. Each provides both a config-driven entry point (run(cfg)) and a programmatic entry point (run_with_data(cfg, dataset)).

Train¶

from dnanet.tasks.train import run, run_with_data

run(cfg) — Full training pipeline from Hydra config. Loads data, builds model, trains, saves checkpoint.

run_with_data(cfg, dataset) — Training with a pre-loaded dataset. Useful for notebooks and testing.

Returns: (trainer, module) tuple for programmatic access.

Evaluate¶

from dnanet.tasks.evaluate import run, run_with_data

run(cfg) — Load checkpoint, run predictions, compute metrics, save results.

run_with_data(cfg, dataset) — Evaluation with a pre-loaded dataset.

Returns: dict[str, float] — metric name → value.

Cross-Validate¶

from dnanet.tasks.cross_validate import run, run_with_data

run(cfg) — K-fold cross-validation from config.

run_with_data(cfg, dataset) — Cross-validation with a pre-loaded dataset.

Returns: {"per_fold": [...], "aggregate": {...}} with mean ± std for each metric.

CLI¶

from dnanet.cli import main

Hydra-decorated main function. Dispatches to the appropriate task runner based on cfg.task.

dnanet task=train data=dnanet_rd model=unet training=segmentation
dnanet task=evaluate checkpoint=best.ckpt
dnanet task=cross_validate training.n_folds=5