Calculations

class lib.gui.analysis.stats.Calculations(session_id, display: str = 'loss', loss_keys: list[str] | str = 'loss', selections: list[str] | str = 'raw', avg_samples: int = 500, smooth_amount: float = 0.9, flatten_outliers: bool = False)

Bases: object

Class that performs calculations on the GlobalSession raw data for the given session id.

Parameters:

session_id (int or None) – The session id number for the selected session from the Analysis tab. Should be None if all sessions are being calculated
display ({"loss", "rate"}, optional) – Whether to display a graph for loss or training rate. Default: “loss”
loss_keys (list, optional) – The list of loss keys to display on the graph. Default: [“loss”]
selections (list, optional) – The selected annotations to display. Default: [“raw”]
avg_samples (int, optional) – The number of samples to use for performing moving average calculation. Default: 500.
smooth_amount (float, optional) – The amount of smoothing to apply for performing smoothing calculation. Default: 0.9.
flatten_outliers (bool, optional) – True if values significantly away from the average should be excluded, otherwise False. Default: False

Attributes Summary

`iterations`	The number of iterations in the data set.
`start_iteration`	The starting iteration number of a limit has been set on the amount of data.
`stats`	The final calculated statistics

Methods Summary

`refresh`()	Refresh the stats
`set_iterations_limit`(limit)	Set the number of iterations to display in the calculations.
`set_smooth_amount`(amount)	Set the amount of smoothing to apply to smoothed graph.
`update_selections`(selection, option)	Update the type of selected data.

Attributes Documentation

iterations

The number of iterations in the data set.

Type:: int

start_iteration

The starting iteration number of a limit has been set on the amount of data.

Type:: int

stats

The final calculated statistics

Type:: dict

Methods Documentation

refresh() → Calculations | None

Refresh the stats

Return type:: Calculations | None

set_iterations_limit(limit: int) → None

Set the number of iterations to display in the calculations.

If a value greater than 0 is passed, then the latest iterations up to the given limit will be calculated.

Parameters:: limit (int) – The number of iterations to calculate data for. 0 to calculate for all data
Return type:: None

set_smooth_amount(amount: float) → None

Set the amount of smoothing to apply to smoothed graph.

Parameters:: amount (float) – The amount of smoothing to apply to smoothed graph
Return type:: None

update_selections(selection: str, option: bool) → None

Update the type of selected data.

Parameters:

selection (str) – The selection to update (as can exist in _selections)
option (bool) – True if the selection should be included, False if it should be removed

Return type:

None

property iterations: int

The number of iterations in the data set.

Type:: int

refresh() → Calculations | None

Refresh the stats

Return type:: Calculations | None

set_iterations_limit(limit: int) → None

Set the number of iterations to display in the calculations.

If a value greater than 0 is passed, then the latest iterations up to the given limit will be calculated.

Parameters:: limit (int) – The number of iterations to calculate data for. 0 to calculate for all data
Return type:: None

set_smooth_amount(amount: float) → None

Set the amount of smoothing to apply to smoothed graph.

Parameters:: amount (float) – The amount of smoothing to apply to smoothed graph
Return type:: None

property start_iteration: int

The starting iteration number of a limit has been set on the amount of data.

Type:: int

property stats: dict[str, ndarray]

The final calculated statistics

Type:: dict

update_selections(selection: str, option: bool) → None

Update the type of selected data.

Parameters:

selection (str) – The selection to update (as can exist in _selections)
option (bool) – True if the selection should be included, False if it should be removed

Return type:

None