utils

Utility functions for graphnet.training.

graphnet.training.utils.collate_fn(graphs)

Remove graphs with less than two DOM hits.

Should not occur in “production”.

Return type:

Batch

Parameters:

graphs (List[Data])

class graphnet.training.utils.collator_sequence_buckleting(batch_splits=[0.8])

Bases: object

Perform the sequence bucketing for the graphs in the batch.

Set cutting points of the different mini-batches.

batch_splits: list of floats, each element is the fraction of the total number of graphs. This list should not explicitly define the first and last elements, which will always be 0 and 1 respectively.

Parameters:

batch_splits (List[float])

graphnet.training.utils.make_dataloader(db, pulsemaps, graph_definition, features, truth, *, batch_size, shuffle, selection, num_workers, persistent_workers, node_truth, truth_table, node_truth_table, string_selection, loss_weight_table, loss_weight_column, index_column, labels)

Construct DataLoader instance.

Return type:

DataLoader

Parameters:

• db (str)

• pulsemaps (str | List[str])

• graph_definition (GraphDefinition)

• features (List[str])

• truth (List[str])

• batch_size (int)

• shuffle (bool)

• selection (List[int] | None)

• num_workers (int)

• persistent_workers (bool)

• node_truth (List[str] | None)

• truth_table (str)

• node_truth_table (str | None)

• string_selection (List[int] | None)

• loss_weight_table (str | None)

• loss_weight_column (str | None)

• index_column (str)

• labels (Dict[str, Callable] | None)

graphnet.training.utils.make_train_validation_dataloader(db, graph_definition, selection, pulsemaps, features, truth, *, batch_size, database_indices, seed, test_size, num_workers, persistent_workers, node_truth, truth_table, node_truth_table, string_selection, loss_weight_column, loss_weight_table, index_column, labels)

Construct train and test DataLoader instances.

Return type:

Tuple[DataLoader, DataLoader]

Parameters:

• db (str)

• graph_definition (GraphDefinition)

• selection (List[int] | None)

• pulsemaps (str | List[str])

• features (List[str])

• truth (List[str])

• batch_size (int)

• database_indices (List[int] | None)

• seed (int)

• test_size (float)

• num_workers (int)

• persistent_workers (bool)

• node_truth (str | None)

• truth_table (str)

• node_truth_table (str | None)

• string_selection (List[int] | None)

• loss_weight_column (str | None)

• loss_weight_table (str | None)

• index_column (str)

• labels (Dict[str, Callable] | None)

graphnet.training.utils.get_predictions(trainer, model, dataloader, prediction_columns, *, node_level, additional_attributes)

Get model predictions on dataloader.

Return type:

DataFrame

Parameters:

• trainer (Trainer)

• model (Model)

• dataloader (DataLoader)

• prediction_columns (List[str])

• node_level (bool)

• additional_attributes (List[str] | None)

graphnet.training.utils.save_results(db, tag, results, archive, model)

Save trained model and prediction results in db.

Return type:

None

Parameters:

• db (str)

• tag (str)

• results (DataFrame)

• archive (str)

• model (Model)

graphnet.training.utils.save_selection(selection, file_path)

Save the list of event numbers to a CSV file.

Parameters:

• selection (List[int]) – List of event ids.

• file_path (str) – File path to save the selection.

Return type:

None

graphnet.training.utils.add_custom_labels(data, custom_label_functions, repeat_labels_by)

Add custom labels to the data.

Parameters:

• data (Data) – data where the label will be stored

• custom_label_functions (Dict[str, Callable[..., Any]]) – dictionary containing the custom label functions

• repeat_labels_by (Optional[int], default: None) – If specified, repeats the labels along the specified dimension.

Return type:

Data

Returns:

data with labels

graphnet.training.utils.add_truth(data, truth_dicts, dtype, repeat_labels_by)

Add truth labels from ´truth_dicts´ to ´data´.

I.e. ´data[key] = truth_dict[key]´

Parameters:

• data (Data) – data where the label will be stored

• truth_dicts (Union[Dict[str, Any], List[Dict[str, Any]]]) – dictionary containing the labels

• dtype (Optional[dtype], default: torch.float32) – dtype of the truth labels

• repeat_labels_by (Optional[int], default: None) – If specified, repeats the labels along the specified dimension.

Return type:

Data

Returns:

data with labels