OakEmu-RL v0.1.0

Appearance

ModelDiagnosticsReport ​

python
ModelDiagnosticsReport(observations: int, episodes: int, wins: int, over_budget: int, trained_activations: list[LayerActivationStatistics], fresh_activations: list[LayerActivationStatistics], trained_weights: list[LayerWeightStatistics], fresh_weights: list[LayerWeightStatistics]) -> None

Activation and weight statistics for a trained model, each beside a freshly initialised model of the same shape.

Source

Methods ​

to_markdown ​

python
to_markdown(title: str) -> str

Render the report as Markdown tables.

Parameters:

  • title (str) – The document title, for example the checkpoint measured.

Returns:

  • str – The Markdown.

Source

Values ​

episodes ​

python
episodes: int

Episodes those observations came from.

Source

fresh_activations ​

python
fresh_activations: list[LayerActivationStatistics]

Activation statistics of the fresh model over the same observations, in the same order.

Source

fresh_weights ​

python
fresh_weights: list[LayerWeightStatistics]

Weight statistics of the fresh model, in the same order.

Source

observations ​

python
observations: int

Observations the activation statistics were measured over.

Source

over_budget ​

python
over_budget: int

How many of those episodes were abandoned at the action budget.

Source

trained_activations ​

python
trained_activations: list[LayerActivationStatistics]

Activation statistics of the trained model, in forward order.

Source

trained_weights ​

python
trained_weights: list[LayerWeightStatistics]

Weight statistics of the trained model, in definition order.

Source

wins ​

python
wins: int

How many of those episodes were won.

Source