lczerolens.lenses#

Lenses module.

Submodules#

Classes#

`ActivationLens`	Class for activation-based XAI methods.
`CompositeLens`	Composite lens for XAI.
`GradientLens`	Class for gradient-based XAI methods.
`LrpLens`	Class for wrapping the LCZero models.
`PatchingLens`	Class for activation-based XAI methods.
`ProbingLens`	Class for probing-based XAI methods.

Package Contents#

class lczerolens.lenses.ActivationLens(pattern=None)[source]#

Bases: lczerolens.lens.Lens

Class for activation-based XAI methods.

Examples

model = LczeroModel.from_path(model_path)
lens = ActivationLens()
board = LczeroBoard()
results = lens.analyse(board, model=model)

Parameters:: pattern (Optional[str])

_intervene(model, **kwargs)[source]#

Intervene on the model.

Parameters:: model (LczeroModel) – The NNsight model.
Returns:: The intervention results.
Return type:: dict

class lczerolens.lenses.CompositeLens(lenses, merge_results=True)[source]#

Bases: lczerolens.lens.Lens

Composite lens for XAI.

Examples

model = LczeroModel.from_path(model_path)
lens = CompositeLens([ActivationLens(), GradientLens()])
board = LczeroBoard()
results = lens.analyse(board, model=model)

Parameters:

lenses (Union[List[lczerolens.lens.Lens], Dict[str, lczerolens.lens.Lens]])
merge_results (bool)

_lens_map#

merge_results = True#

is_compatible(model)[source]#

Returns whether the lens is compatible with the model.

Parameters:: model (LczeroModel) – The NNsight model.
Returns:: Whether the lens is compatible with the model.
Return type:: bool

prepare(model, **kwargs)[source]#

Prepare the model for the lens.

Parameters:: model (LczeroModel) – The NNsight model.
Returns:: The prepared model.
Return type:: LczeroModel

_intervene(model, **kwargs)[source]#

Intervene on the model.

Parameters:: model (LczeroModel) – The NNsight model.
Returns:: The intervention results.
Return type:: dict

class lczerolens.lenses.GradientLens(*, input_requires_grad=True, **kwargs)[source]#

Bases: lczerolens.lens.Lens

Class for gradient-based XAI methods.

Parameters:: input_requires_grad (bool)

input_requires_grad = True#

_intervene(model, **kwargs)[source]#

Intervene on the model.

Parameters:: model (LczeroModel) – The NNsight model.
Returns:: The intervention results.
Return type:: dict

class lczerolens.lenses.LrpLens(pattern=None)[source]#

Bases: lczerolens.lens.Lens

Class for wrapping the LCZero models.

Parameters:: pattern (Optional[str])

_intervene(model, **kwargs)[source]#

Intervene on the model.

Parameters:: model (LczeroModel) – The NNsight model.
Returns:: The intervention results.
Return type:: dict

class lczerolens.lenses.PatchingLens(patch_fn, **kwargs)[source]#

Bases: lczerolens.lens.Lens

Class for activation-based XAI methods.

Examples

model = LczeroModel.from_path(model_path)
lens = PatchingLens()
board = LczeroBoard()
patch_fn = lambda n, m, *kwargs: pass
results = lens.analyse(board, model=model)

Parameters:: patch_fn (Callable)

_patch_fn#

_intervene(model, **kwargs)[source]#

Intervene on the model.

Parameters:: model (LczeroModel) – The NNsight model.
Returns:: The intervention results.
Return type:: dict

class lczerolens.lenses.ProbingLens(probe_fn, **kwargs)[source]#

Bases: lczerolens.lens.Lens

Class for probing-based XAI methods.

Examples

model = LczeroModel.from_path(model_path)
lens = ProbingLens(probe)
board = LczeroBoard()
results = lens.analyse(board, model=model)

Parameters:: probe_fn (Callable)

_probe_fn#

_intervene(model, **kwargs)[source]#

Intervene on the model.

Parameters:: model (LczeroModel) – The NNsight model.
Returns:: The intervention results.
Return type:: dict