State

class plugins.train.model._base.state.State(model_dir: str, model_name: str, no_logs: bool)

Bases: object

Holds state information relating to the plugin’s saved model.

Parameters:

model_dir (str) – The full path to the model save location
model_name (str) – The name of the model plugin
no_logs (bool) – True if Tensorboard logs should not be generated, otherwise False

Attributes Summary

`current_session`	The state dictionary for the current `session_id`.
`filename`	Full path to the state filename
`iterations`	The total number of iterations that the model has trained.
`lr_finder`	The value discovered from the learning rate finder.
`mixed_precision_layers`	Layers that can be switched between mixed-float16 and float32.
`model_needs_rebuild`	`True` if mixed precision policy has changed so model needs to be rebuilt otherwise `False`
`session_id`	The current training session id.
`sessions`	The session information for each session in the state file

Methods Summary

`add_lr_finder`(learning_rate)	Add the optimal discovered learning rate from the learning rate finder
`add_mixed_precision_layers`(layers)	Add the list of model's layers that are compatible for mixed precision to the state dictionary
`add_session_batchsize`(batch_size)	Add the session batch size to the sessions dictionary.
`increment_iterations`()	Increment `iterations` and session iterations by 1.
`save`()	Save the state values to the serialized state file.
`update_session_config`(key, value)	Update a configuration item of the currently loaded session.

Attributes Documentation

current_session

The state dictionary for the current session_id.

Type:: dict

filename

Full path to the state filename

Type:: str

iterations

The total number of iterations that the model has trained.

Type:: int

lr_finder: The value discovered from the learning rate finder. -1 if no value stored

mixed_precision_layers

Layers that can be switched between mixed-float16 and float32.

Type:: list

model_needs_rebuild

True if mixed precision policy has changed so model needs to be rebuilt otherwise False

Type:: bool

session_id

The current training session id.

Type:: int

sessions

The session information for each session in the state file

Type:: dict[int, dict[str, Any]]

Methods Documentation

add_lr_finder(learning_rate: float) → None

Add the optimal discovered learning rate from the learning rate finder

Parameters:: learning_rate (float) – The discovered learning rate
Return type:: None

add_mixed_precision_layers(layers: list[str]) → None

Add the list of model’s layers that are compatible for mixed precision to the state dictionary

Parameters:: layers (list[str])
Return type:: None

add_session_batchsize(batch_size: int) → None

Add the session batch size to the sessions dictionary.

Parameters:: batch_size (int) – The batch size for the current training session
Return type:: None

increment_iterations() → None

Increment iterations and session iterations by 1.

Return type:: None

save() → None

Save the state values to the serialized state file.

Return type:: None

update_session_config(key: str, value: Any) → None

Update a configuration item of the currently loaded session.

Parameters:

key (str) – The configuration item to update for the current session
value (any) – The value to update to

Return type:

None

add_lr_finder(learning_rate: float) → None

Add the optimal discovered learning rate from the learning rate finder

Parameters:: learning_rate (float) – The discovered learning rate
Return type:: None

add_mixed_precision_layers(layers: list[str]) → None

Add the list of model’s layers that are compatible for mixed precision to the state dictionary

Parameters:: layers (list[str])
Return type:: None

add_session_batchsize(batch_size: int) → None

Add the session batch size to the sessions dictionary.

Parameters:: batch_size (int) – The batch size for the current training session
Return type:: None

property current_session: dict

The state dictionary for the current session_id.

Type:: dict

property filename: str

Full path to the state filename

Type:: str

increment_iterations() → None

Increment iterations and session iterations by 1.

Return type:: None

property iterations: int

The total number of iterations that the model has trained.

Type:: int

lowest_avg_loss: float

The lowest average loss seen between save intervals.

Type:: float

property lr_finder: float: The value discovered from the learning rate finder. -1 if no value stored

property mixed_precision_layers: list[str]

Layers that can be switched between mixed-float16 and float32.

Type:: list

property model_needs_rebuild: bool

True if mixed precision policy has changed so model needs to be rebuilt otherwise False

Type:: bool

save() → None

Save the state values to the serialized state file.

Return type:: None

property session_id: int

The current training session id.

Type:: int

property sessions: dict[int, dict[str, Any]]

The session information for each session in the state file

Type:: dict[int, dict[str, Any]]

update_session_config(key: str, value: Any) → None

Update a configuration item of the currently loaded session.

Parameters:

key (str) – The configuration item to update for the current session
value (any) – The value to update to

Return type:

None