drytorch.trackers.logging

Module containing custom logging configurations for the ‘drytorch’ logger.

It defines and implements a formatter that formats log messages according to the levels defined in the INFO_LEVELS variable. By default, it prints to stream and does not propagate to the main root.

INFO_LEVELS: InfoLevels object for global settings.

Functions

`disable_default_handler`()	Disable the handler and filter of the local logger.
`disable_propagation`()	Revert the changes made by enable_propagation.
`enable_default_handler`(stream)	Set up the default logging configuration.
`enable_propagation`([deduplicate_stream])	Propagate to the root logger.
`get_verbosity`()	Get the verbosity level of the 'drytorch' logger.
`set_formatter`(style)	Set the formatter for the stream handler of the drytorch logger.
`set_verbosity`(level_no)	Set the verbosity level of the 'drytorch' logger.

Classes

`BuiltinLogger`()	Tracker that streams logging messages through the built-in logger.
`DryTorchFilter`()	Filter that excludes logs from 'drytorch'.
`DryTorchFormatter`()	Default formatter for the drytorch logger.
`InfoLevels`(internal, metrics, epoch, ...)	Dataclass that defines different levels of information for logging.
`ProgressFormatter`()	Formatter that dynamically overwrites metrics and epoch logs.

class BuiltinLogger[source]

Bases: Tracker

Tracker that streams logging messages through the built-in logger.

notify(event: Event) → None[source]

notify(event: StartTrainingEvent) → None

notify(event: EndTrainingEvent) → None

notify(event: StartEpochEvent) → None

notify(event: EndEpochEvent) → None

notify(event: SaveModelEvent) → None

notify(event: LoadModelEvent) → None

notify(event: MetricEvent) → None

notify(event: StartTestEvent) → None

notify(event: EndTestEvent) → None

notify(event: TerminatedTrainingEvent) → None

notify(event: StartExperimentEvent) → None

notify(event: StopExperimentEvent) → None

notify(event: LearningRateEvent) → None

notify(event: ModelRegistrationEvent) → None

notify(event: ActorRegistrationEvent) → None

Notify the tracker of an event.

Parameters:: event (Event) – the event to notify about.
Return type:: None

class DryTorchFilter[source]

Bases: Filter

Filter that excludes logs from ‘drytorch’.

Initialize.

filter(record: LogRecord) → bool[source]

Filter logs propagated by the library logger.

Parameters:: record (LogRecord)
Return type:: bool

class DryTorchFormatter[source]

Bases: Formatter

Default formatter for the drytorch logger.

default_msec_format

format for milliseconds.

Type:: ClassVar[str]

Initialize.

format(record: LogRecord) → str[source]

Format the log record.

Parameters:: record (LogRecord)
Return type:: str

class InfoLevels(internal: int, metrics: int, epoch: int, model_state: int, experiment: int, training: int, test: int)[source]

Bases: object

Dataclass that defines different levels of information for logging.

Parameters:

internal (int)
metrics (int)
epoch (int)
model_state (int)
experiment (int)
training (int)
test (int)

internal

level for internal logging messages.

Type:: int

metrics

level for metric reporting.

Type:: int

epoch

level for epoch-related messages.

Type:: int

model_state

level for model state changes.

Type:: int

experiment

level for experiment-related messages.

Type:: int

training

level for training-related messages.

Type:: int

test

level for test-related messages.

Type:: int

class ProgressFormatter[source]

Bases: DryTorchFormatter

Formatter that dynamically overwrites metrics and epoch logs.

Initialize.

disable_default_handler() → None[source]

Disable the handler and filter of the local logger.

Return type:: None

disable_propagation() → None[source]

Revert the changes made by enable_propagation.

Return type:: None

enable_default_handler(stream: SupportsWrite[str] = <_io.TextIOWrapper name='<stderr>' mode='w' encoding='utf-8'>) → None[source]

Set up the default logging configuration.

Parameters:: stream (SupportsWrite[str])
Return type:: None

enable_propagation(deduplicate_stream: bool = True) → None[source]

Propagate to the root logger.

Parameters:: deduplicate_stream (bool) – whether to remove local messages from the stream.
Return type:: None

get_verbosity() → int[source]

Get the verbosity level of the ‘drytorch’ logger.

Return type:: int

set_formatter(style: Literal['drytorch', 'progress']) → None[source]

Set the formatter for the stream handler of the drytorch logger.

Raises:: ValueError – if the style is not ‘drytorch’ or ‘progress’.
Parameters:: style (Literal['drytorch', 'progress'])
Return type:: None

set_verbosity(level_no: int)[source]

Set the verbosity level of the ‘drytorch’ logger.

Parameters:: level_no (int)