fnet.cli package¶
Submodules¶
fnet.cli.init module¶
-
fnet.cli.init.add_parser_arguments(parser: argparse.ArgumentParser) → None[source]¶ Add init script arguments to parser.
-
fnet.cli.init.main(args: Optional[argparse.Namespace] = None) → None[source]¶ Install default training options and example model training/prediction scripts into current directory.
-
fnet.cli.init.save_default_predict_options(path_save: pathlib.Path) → None[source]¶ Save default prediction options json.
- Parameters
path_save – Save path for default prediction options json.
-
fnet.cli.init.save_default_train_options(path_save: pathlib.Path) → None[source]¶ Save default training options json.
- Parameters
path_save – Save path for default training options json.
fnet.cli.main module¶
Module for command-line ‘fnet’ command.
fnet.cli.predict module¶
Generates predictions from a model.
-
fnet.cli.predict.add_parser_arguments(parser) → None[source]¶ Add training script arguments to parser.
-
fnet.cli.predict.get_dataset(args: argparse.Namespace) → torch.utils.data.dataset.Dataset[source]¶ Returns dataset.
- Returns
Dataset object.
- Return type
torch.utils.data.Dataset
-
fnet.cli.predict.get_indices(args: argparse.Namespace, dataset: Any) → List[int][source]¶ Returns indices of dataset items on which to perform predictions.
-
fnet.cli.predict.item_from_dataset(dataset: Any, idx: int) → Tuple[torch.Tensor, Optional[torch.Tensor]][source]¶ Returns signal-target image pair from dataset.
If the dataset is a FnetDataset, it will be indexed using ‘loc’-style indexing.
- Parameters
dataset – Object with __getitem__ implemented.
idx – Index of data to be retrieved from dataset.
- Returns
Signal-target data pair. Target can be None if dataset does not return a target for the given index.
- Return type
Tuple[torch.Tensor, Optional[torch.Tensor]]
-
fnet.cli.predict.load_from_json(args: argparse.Namespace) → None[source]¶ Loads arguments from if a json is specified.
-
fnet.cli.predict.main(args: Optional[argparse.Namespace] = None) → None[source]¶ Predicts using model.
-
fnet.cli.predict.parse_model(model_str: str) → Dict[source]¶ Parse model definition string into dictionary.
-
fnet.cli.predict.save_args_as_json(path_save_dir: str, args: argparse.Namespace) → None[source]¶ Saves script arguments as json in save directory.
A json is saved only if the “–json” option was not specified.
By default, this function tries to save arguments as predict_options.json within the save directory. If that file already exists, appends a digit to uniquify the save path.
- Parameters
path_save_dir – Save directory
args – Script arguments.
-
fnet.cli.predict.save_predictions_csv(path_csv: pathlib.Path, pred_records: List[Dict], dataset: Any) → None[source]¶ Saves csv with metadata of predictions.
- Parameters
path_csv – CSV save path.
pred_records – List of metadata for each prediction.
dataset – Dataset from where signal-target pairs were retrieved.
-
fnet.cli.predict.save_tif(fname: str, ar: numpy.ndarray, path_root: str) → str[source]¶ Saves a tif and returns tif save path relative to root save directory.
Image will be stored at: ‘path_root/tifs/fname’
- Parameters
fname – Basename of save path.
ar – Array to be saved as tif.
path_root – Root directory of save path.
- Returns
Save path relative to root directory.
- Return type
str
fnet.cli.train_model module¶
Trains a model.
-
fnet.cli.train_model.add_parser_arguments(parser) → None[source]¶ Add training script arguments to parser.
-
fnet.cli.train_model.get_bpds_train(args: argparse.Namespace) → fnet.data.bufferedpatchdataset.BufferedPatchDataset[source]¶ Creates data provider for training.
-
fnet.cli.train_model.get_bpds_val(args: argparse.Namespace) → Optional[fnet.data.bufferedpatchdataset.BufferedPatchDataset][source]¶ Creates data provider for validation.
-
fnet.cli.train_model.train_model(batch_size: int = 28, bpds_kwargs: Optional[Dict] = None, dataset_train: str = 'fnet.data.TiffDataset', dataset_train_kwargs: Optional[Dict] = None, dataset_val: Optional[str] = None, dataset_val_kwargs: Optional[Dict] = None, fnet_model_class: str = 'fnet.fnet_model.Model', fnet_model_kwargs: Optional[Dict] = None, interval_checkpoint: int = 50000, interval_save: int = 1000, iter_checkpoint: Optional[List] = None, n_iter: int = 250000, path_save_dir: str = 'models/some_model', seed: Optional[int] = None, json: Optional[str] = None, gpu_ids: Optional[List[int]] = None)[source]¶ Python API for training.