Logger

minnt.Logger

Bases: ABC

An abstract logger interface for logging metrics and other information.

close abstractmethod

close() -> None

Close the logger and release its resources.

log_audio abstractmethod

log_audio(label: str, audio: AnyArray, sample_rate: int, epoch: int) -> Self

Log the given audio with the given label at the given epoch.

Parameters:

• label (str) –

  The label of the logged audio.

• audio (AnyArray) –

  The audio to log, represented as an array with any of the following shapes:

  • (L,) of (L, 1) for mono audio,
  • (L, 2) for stereo audio.

  If the sample values are floating-point numbers, they are expected to be in the [-1, 1] range; otherwise, they are assumed to be in the [-32_768, 32_767] range.

• sample_rate (int) –

  The sample rate of the audio.

• epoch (int) –

  The epoch number at which the audio is logged.

log_config abstractmethod

log_config(config: dict[str, Any], epoch: int) -> Self

Log the given configuration dictionary at the given epoch.

Parameters:

• config (dict[str, Any]) –

  A JSON-serializable dictionary representing the configuration to log.

• epoch (int) –

  The epoch number at which the configuration is logged.

log_figure abstractmethod

log_figure(
    label: str,
    figure: Any,
    epoch: int,
    tight_layout: bool = True,
    close: bool = True,
) -> Self

Log the given matplotlib Figure with the given label at the given epoch.

Parameters:

• label (str) –

  The label of the logged image.

• figure (Any) –

  A matplotlib Figure.

• epoch (int) –

  The epoch number at which the image is logged.

• tight_layout (bool, default: True ) –

  Whether to apply tight layout to the figure before logging it.

• close (bool, default: True ) –

  Whether to close the figure after logging it.

log_graph abstractmethod

log_graph(graph: Module, data: TensorOrTensors, epoch: int) -> Self

Log the given computation graph by tracing it with the given data.

Alternatively, loggers may choose to log the graph using TorchScript, run it on the given data, or use any other mechanism they see fit.

Parameters:

• graph (Module) –

  The computation graph to log, represented as a PyTorch module.

• data (TensorOrTensors) –

  The input data to use for tracing the computation graph.

• epoch (int) –

  The epoch number at which the computation graph is logged.

log_image abstractmethod

log_image(
    label: str, image: AnyArray, epoch: int, data_format: DataFormat = "HWC"
) -> Self

Log the given image with the given label at the given epoch.

Parameters:

• label (str) –

  The label of the logged image.

• image (AnyArray) –

  The image to log, represented as a PIL image or as an array of any of the following shapes (assuming "HWC" data format):

  • (H, W) or (H, W, 1) for grayscale images,
  • (H, W, 2) for grayscale images with alpha channel,
  • (H, W, 3) for RGB images,
  • (H, W, 4) for RGBA images.

  If the pixel values are floating-point numbers, they are expected to be in the [0, 1] range; otherwise, they are assumed to be in the [0, 255] range.

• epoch (int) –

  The epoch number at which the image is logged.

• data_format (DataFormat, default: 'HWC' ) –

  The data format of the image specifying whether the channels are stored in the last dimension ("HWC", the default) or in the first dimension ("CHW"); ignored for a PIL image.

log_metrics abstractmethod

log_metrics(
    logs: dict[str, float], epoch: int, description: str | None = None
) -> Self

Log metrics collected during a given epoch, with an optional description.

Parameters:

• logs (dict[str, float]) –

  A dictionary of logged metrics for the epoch.

• epoch (int) –

  The epoch number at which the logs were collected.

• description (str | None, default: None ) –

  An optional description of the logged metrics (used only by some loggers).

log_text abstractmethod

log_text(label: str, text: str, epoch: int) -> Self

Log the given text with the given label at the given epoch.

Parameters:

• label (str) –

  The label of the logged text.

• text (str) –

  The text to log.

• epoch (int) –

  The epoch number at which the text is logged.