lucent package

Subpackages

• lucent.misc package
  • misc.io
    • lucent.misc.io package
      • Submodules
      • lucent.misc.io.collapse_channels module
      • lucent.misc.io.serialize_array module
      • lucent.misc.io.showing module
      • Module contents
  • misc.channel_reducer module
  • Module contents
• lucent.optvis package
  • optvis.param
    • lucent.optvis.param package
      • param.color module
      • param.cppn module
      • param.gan module
      • param.images module
      • param.lowres module
      • param.spatial module
      • Module contents
  • optvis.objectives module
  • optvis.objectives_util module
  • optvis.render module
  • optvis.transform module
  • Module contents

lucent.util module

Utility functions

lucent.util.set_seed(seed)

Seeds all random number generators and enables deterministic cudnn. : param seed: seed value to seed the RNGs. Can e.g. be float or int.

lucent.model_utils module

Utility functions for models.

lucent.model_utils.add_to_dependence_graph(dependence_graph, prefix, name)

lucent.model_utils.filter_layer_names(layer_names, depth)

lucent.model_utils.get_model_layers(model: Module, getLayerRepr: Optional[bool] = False, dependencies: Optional[bool] = True, excludeNorms: Optional[bool] = True, excludeActs: Optional[bool] = True, excludePools: Optional[bool] = True) → Union[List[str], OrderedDict]

Get the names of all layers of a network. The names are given in the format that can be used

to access them via objectives.

Parameters

• model (torch.nn.Module) – the network to get the names of

• getLayerRepr (Optional[bool], optional) – whether to return a OrderedDict of layer names, layer representation string pair. If False just return a list of names.

• dependencies – Whether to return dependencies of layers as a nested OrderedDict

• excludeNorms (Optional[bool], optional) – whether to exclude normalization layers, defaults to True

• excludeActs (Optional[bool], optional) – whether to exclude Activation layers, defaults to True

• excludePools (Optional[bool], optional) – whether to exclude Pooling layers, defaults to True

Rtype dependencies

Optional[bool], optional

Raises

• ValueError – model has wrong type

• ValueError – model has no modules

Returns

dict of name, repr pairs or just list of names of all layers (including activations if they are instantiated as layers)

Return type

Union[List[str], OrderedDict]

Module contents