mltk.core.WeightsAndBiasesMixin

class WeightsAndBiasesMixin

Provides various properties to the base MltkModel used by Weights & Biases 3rd-party cloud backend.

See also

• Tutorial: Cloud logging with Weights & Biases

• Weights & Biases Documentation

Properties

wandb_callback	Keras callback to automatically log info to wandb
wandb_config	Additional configuration values
wandb_init_kwargs	Additional arguments to provide to wandb.init()
wandb_is_disabled	Manually disable the wandb backend
wandb_is_initialized	Return if the wandb backend is initialized
wandb_model_checkpoint_callback	Callback to periodically save the Keras model or model weights
wandb_session_id	The wandb project session or run ID

Methods

__init__
wandb_log	Logs a dictionary of data to the current wandb run's history
wandb_save	Save files to wandb cloud

property wandb_is_initialized

Return if the wandb backend is initialized

property wandb_is_disabled

Manually disable the wandb backend

property wandb_init_kwargs

Additional arguments to provide to wandb.init()

The following argument are automatically populated by this mixin:

• project - The name of the name

• job_type - train, evaluation, quantize, or profile

• dir - The log_dir

• id - The timestamp when the training was invoked. See wandb_session_id This id is re-used by the evaluate, profile, and quantize commands

• resume - Set to never for the train command, and must otherwise

See wandb.init() for the other available arguments

property wandb_config

Additional configuration values

This sets the wandb.config object in your script to save your training configuration: hyperparameters, input settings like dataset name or model type, and any other independent variables for your experiments. This is useful for analyzing your experiments and reproducing your work in the future. You’ll be able to group by config values in the web interface, comparing the settings of different runs and seeing how these affect the output.

property wandb_callback

Keras callback to automatically log info to wandb

This allows for specifying a custom wandb.keras.WandbCallback. If not set, then the mixin will automatically populate this callback.

property wandb_model_checkpoint_callback

Callback to periodically save the Keras model or model weights

This allows for specifying a custom wandb.keras.WandbModelCheckpoint.

property wandb_session_id

The wandb project session or run ID

This is the timestamp of when the last train command was invoked for the model. This ID is re-used for evaluate, profile, and quantize commands.

This value is used at the id argument to wandb.init()

wandb_save(glob_str=None, base_path=None, policy='live', logger=None)

Save files to wandb cloud

Internally, this invokes wandb.save()

Parameters:

• glob_str (Optional[str]) – a relative or absolute path to a unix glob or regular path. If this isn’t specified the method is a noop.

• base_path (Optional[str]) – the base path to run the glob relative to

• policy –

  one of live, now, or end

  • live: upload the file as it changes, overwriting the previous version

  • now: upload the file once now

  • end: only upload file when the run ends

• logger (Logger) –

wandb_log(data, step=None, commit=True, logger=None)

Logs a dictionary of data to the current wandb run’s history

Internally, this invokes wandb.log()

Parameters:

• data (Dict[str, Any]) – A dict of serializable python objects i.e str, ints, floats, Tensors, dicts, or any of the wandb.data_types

• commit (Optional[bool]) – Save the metrics dict to the wandb server and increment the step. If false wandb.log just updates the current metrics dict with the data argument and metrics won’t be saved until wandb.log is called with commit=True.

• step (Optional[int]) – The global step in processing. This persists any non-committed earlier steps but defaults to not committing the specified step.

• logger (Logger) –