metallic.utils

Logger

class metallic.utils.Logger(root: str, n_iters_per_epoch: int, log_basename: str = '', log_interval: int = 100, tensorboard: bool = False, verbose: bool = False)

Bases: object

A loggger to log and visualize (based on Tensorboard) statistics.

Parameters

• root (str) – Root directory of the log files

• n_iters_per_epoch (int) – Number of the iterations per epoch

• log_basename (str, optional, default='') – Base name of the log file

• log_interval (int, optional, default=100) – Steps between info loggings

• tensorboard (bool, optional, default=False) – Enable tensorboard or not (tensorboard package is required)

• verbose (bool, optional, default=False) –

log(data: dict, epoch: int, i_iter: int, stage: str) → None

Log statistics generated during updating.

Parameters

• data (dict) – Data to be logged

• epoch (int) – Epoch of the data to be logged

• i_iter (int) – Iteration of the data to be logged

• stage (str) – Name of the current stage

write_tensorboard(key: str, x: numbers.Number, y: numbers.Number) → None

Log data into Tensorboard.

Parameters

• key (str) – Namespace which the input data tuple belongs to

• x (Union[Number, np.number]) – Ordinate of the input data

• y (Union[Number, np.number]) – Abscissa of the input data

write_text(text: str) → None

Log data into text files.

Parameters

text (str) – A string to be logged

Metrics

class metallic.utils.MetricTracker(*names)

Bases: object

Keep track of metrics.

add(name: str) → None

Add a new metric.

property metrics

Return a dict containing all metrics.

reset() → None

Clear all of the recorded metrics.

update(name: str, value: numbers.Number) → None

Update a new value to a specified metric.

metallic.utils.get_accuracy(scores: torch.Tensor, targets: torch.Tensor) → float

Compute accuracy using predicted scores and targets.