noether.core.trackers¶

Submodules¶

Classes¶

`BaseTracker`	Base class for all experiment trackers.
`NoopTracker`	Dummy tracker that does nothing.
`TensorboardTracker`	TensorBoard tracker for logging metrics and configuration.
`TrackioTracker`	HuggingFace Trackio tracker.
`WandBTracker`	Weights and Biases tracker.

Package Contents¶

class noether.core.trackers.BaseTracker(metric_property_provider, path_provider)¶

Base class for all experiment trackers.

Initialize the BaseTracker.

Parameters:

metric_property_provider (noether.core.providers.MetricPropertyProvider) – The metric property provider gives additional information such as whether higher values are better.
path_provider (noether.core.providers.PathProvider) – Gives access to paths (e.g., output_path: where checkpoints/logs are stored).

logger¶

metric_property_provider¶

path_provider¶

summary: dict[str, Any]¶

init(accelerator, run_name, stage_name, stage_hp, run_id, output_uri)¶

Initialize the tracker for a specific run.

Parameters:

accelerator (str) – The accelerator used for training (e.g., “cpu”, “cuda”).
run_name (str) – The name of the run.
stage_name (str | None) – The stage of the run.
stage_hp (dict) – The hyperparameters for the run stage.
run_id (str) – The ID of the run.
output_uri (str) – The URI where the output will be stored.

Return type:

None

log(data)¶

Log data to the tracker.

Parameters:: data (dict[str, Any]) – The data to log. This should be a dictionary where the values are the data to log.
Return type:: None

close()¶

Close the file used by the tracker and save the summary.

Return type:: None

set_summary(key, value)¶

Set a summary value. :param key: The key for the summary value. :param value: The value to set.

Parameters:

key (str)
value (Any)

Return type:

None

update_summary(data)¶

Update the summary with new data.

Parameters:: data (dict) – The data to update the summary with. This should be a dictionary where the values are the data to log.

summarize_logvalues()¶

Summarize the log values from the entries. This method is called after the training is finished and summarizes the log values. It computes the min/max values for each log value and stores them in the summary.

Return type:: None

class noether.core.trackers.NoopTracker(metric_property_provider, path_provider)¶

Bases: noether.core.trackers.base.BaseTracker

Dummy tracker that does nothing.

Initialize the BaseTracker.

Parameters:

metric_property_provider (noether.core.providers.MetricPropertyProvider) – The metric property provider gives additional information such as whether higher values are better.
path_provider (noether.core.providers.PathProvider) – Gives access to paths (e.g., output_path: where checkpoints/logs are stored).

class noether.core.trackers.TensorboardTracker(tracker_config, **kwargs)¶

Bases: noether.core.trackers.base.BaseTracker

TensorBoard tracker for logging metrics and configuration.

Initialize the TensorboardTracker.

Parameters:

tracker_config (noether.core.schemas.trackers.TensorboardTrackerSchema) – Configuration for TensorBoard. See TensorboardTrackerSchema for available options (typically including base log_dir).
**kwargs – Additional keyword arguments passed to the parent class.

config¶

writer: torch.utils.tensorboard.SummaryWriter | None = None¶

class noether.core.trackers.TrackioTracker(tracker_config, **kwargs)¶

Bases: noether.core.trackers.base.BaseTracker

HuggingFace Trackio tracker.

https://github.com/gradio-app/trackio

Initialize the TrackioTracker. :param tracker_config: Configuration for the TrackioTracker. See TrackioTrackerSchema for available options including project and space_id. :param **kwargs: Additional keyword arguments passed to the parent class.

Parameters:: tracker_config (noether.core.schemas.trackers.TrackioTrackerSchema)

config¶

class noether.core.trackers.WandBTracker(tracker_config, **kwargs)¶

Bases: noether.core.trackers.base.BaseTracker

Weights and Biases tracker.

Example config:

Parameters:

tracker_config (noether.core.schemas.trackers.WandBTrackerSchema) – Configuration for the WandBTracker. See WandBTrackerSchema for available options including mode, entity, and project.
**kwargs – Additional keyword arguments passed to the parent class.

MODES = ['disabled', 'online', 'offline']¶

mode¶

entity¶

project¶