tdhook.attribution.activation_maximisation

Classes

ActivationMaximisation

Activation maximisation [14].

Module Contents

class tdhook.attribution.activation_maximisation.ActivationMaximisation(modules_to_maximise, alpha=0.1, n_steps=10, min_value=-float('Inf'), max_value=float('Inf'), init_attr_targets=None, init_attr_grads=None, additional_init_keys=None, attribution_key='attr', clean_intermediate_keys=True)

Bases: tdhook.contexts.HookingContextFactory

Activation maximisation [14].

Parameters:

• modules_to_maximise (List[str])

• alpha (float)

• n_steps (int)

• min_value (float)

• max_value (float)

• init_attr_targets (Optional[Callable[[tensordict.TensorDict, tensordict.TensorDict], tensordict.TensorDict]])

• init_attr_grads (Optional[Callable[[tensordict.TensorDict, tensordict.TensorDict], tensordict.TensorDict]])

• additional_init_keys (Optional[List[tdhook._types.UnraveledKey]])

• attribution_key (tdhook._types.UnraveledKey)

• clean_intermediate_keys (bool)

_attribution_key = 'attr'

_modules_to_maximise

_alpha = 0.1

_n_steps = 10

_min_value

_max_value

_clean_intermediate_keys = True

_prepare_module(module, in_keys, out_keys, extra_relative_path)

Parameters:

• module (tensordict.nn.TensorDictModuleBase)

• in_keys (List[tdhook._types.UnraveledKey])

• out_keys (List[tdhook._types.UnraveledKey])

• extra_relative_path (str)

Return type:

tensordict.nn.TensorDictModuleBase