GlobalSession

class lib.gui.analysis.stats.GlobalSession

Bases: object

Holds information about a loaded or current training session by accessing a model’s state file and Tensorboard logs. This class should not be accessed directly, rather through lib.gui.analysis.Session

Attributes Summary

`batch_sizes`	The batch sizes for each session_id for the model.
`full_summary`	List of dictionaries containing summary statistics for each session id.
`have_session_data`	`True` if session data is available otherwise `False`
`is_loaded`	`True` if session data is loaded otherwise `False`
`is_training`	`True` if the loaded session is the currently training model, otherwise `False`
`logging_disabled`	`True` if logging is disabled for the currently training session otherwise `False`.
`model_filename`	The full model filename
`session_ids`	The sorted list of all existing session ids in the state file

Methods Summary

`clear`()	Clear the currently loaded session.
`get_loss`(session_id)	Obtain the loss values for the given session_id.
`get_loss_keys`(session_id)	Obtain the loss keys for the given session_id.
`get_timestamps`(-> dict[int, ~numpy.ndarray])	Obtain the time stamps keys for the given session_id.
`initialize_session`(model_folder, model_name)	Initialize a Session.
`stop_training`()	Clears the internal training flag.

Attributes Documentation

batch_sizes

The batch sizes for each session_id for the model.

Type:: dict

full_summary

List of dictionaries containing summary statistics for each session id.

Type:: list

have_session_data

True if session data is available otherwise False

Type:: bool

is_loaded

True if session data is loaded otherwise False

Type:: bool

is_training

True if the loaded session is the currently training model, otherwise False

Type:: bool

logging_disabled

True if logging is disabled for the currently training session otherwise False.

Type:: bool

model_filename

The full model filename

Type:: str

session_ids

The sorted list of all existing session ids in the state file

Type:: list

Methods Documentation

clear() → None

Clear the currently loaded session.

Return type:: None

get_loss(session_id: int | None) → dict[str, ndarray]

Obtain the loss values for the given session_id.

Parameters:: session_id (int or None) – The session ID to return loss for. Pass None to return loss for all sessions.
Returns:: Loss names as key, numpy.ndarray as value. If No session ID was provided all session’s losses are collated
Return type:: dict

get_loss_keys(session_id: int | None) → list[str]

Obtain the loss keys for the given session_id.

Parameters:: session_id (int or None) – The session ID to return the loss keys for. Pass None to return loss keys for all sessions.
Returns:: The loss keys for the given session. If None is passed as session_id then a unique list of all loss keys for all sessions is returned
Return type:: list

get_timestamps(session_id: None) → dict[int, ndarray]

get_timestamps(session_id: int) → ndarray

Obtain the time stamps keys for the given session_id.

Parameters:: session_id (int or None) – The session ID to return the time stamps for. Pass None to return time stamps for all sessions.
Returns:: If a session ID has been given then a single numpy.ndarray will be returned with the session’s time stamps. Otherwise a ‘dict’ will be returned with the session IDs as key with numpy.ndarray of timestamps as values
Return type:: dict[int] or numpy.ndarray

initialize_session(model_folder: str, model_name: str, is_training: bool = False) → None

Initialize a Session.

Load’s the model’s state file, and sets the paths to any underlying Tensorboard logs, ready for access on request.

Parameters:

model_folder (str,) – If loading a session manually (e.g. for the analysis tab), then the path to the model folder must be provided. For training sessions, this should be passed through from the launcher
model_name (str, optional) – If loading a session manually (e.g. for the analysis tab), then the model filename must be provided. For training sessions, this should be passed through from the launcher
is_training (bool, optional) – True if the session is being initialized for a training session, otherwise False. Default: False

Return type:

None

stop_training() → None

Clears the internal training flag. To be called when training completes.

Return type:: None

property batch_sizes: dict[int, int]