Parameters Conversion#

initialize_module(module: torch.nn.modules.module.Module, vec: torch.Tensor, clone=True, detach=True)[source]#

initializes a module's parameters with a 1-D vector

Parameters
  • module (Module) -- module to initialize weights

  • vec (Tensor) -- a 1-D Tensor

  • clone (bool, optional) -- clones the vector before initilization. Defaults to True.

  • detach (bool, optional) -- detaches the output before the initialization. Defaults to True.

vector_to_named_parameters_like(vec: torch.Tensor, named_parameters_like: collections.OrderedDict) collections.OrderedDict[source]#

Convert one vector to new named parameters like the ones provided

Parameters
  • vec (Tensor) -- a single vector represents the parameters of a model.

  • parameters (OrderedDict) -- a dictioanry of Tensors that are the parameters of a model. This is only used to get the sizes and keys. New parametere are defined.

vector_to_parameters_like(vec, parameters_like)[source]#

Convert one vector to new parameters like the ones provided

Parameters
  • vec (Tensor) -- a single vector represents the parameters of a model.

  • parameters (Iterable[Tensor]) -- an iterator of Tensors that are the parameters of a model. This is only used to get the sizes. New parametere are defined.

vectorize_module(module: torch.nn.modules.module.Module, clone=True, detach=True)[source]#

convert parameters of a module to a vector

Parameters
  • module (Module) -- module to convert the parameters of

  • clone (bool, optional) -- clones the output. Defaults to True.

  • detach (bool, optional) -- detaches the output. Defaults to True.

Returns

Module -- 1-D Tensor of all parameters in the module

vectorize_module_grads(module: torch.nn.modules.module.Module, clone=True, detach=True)[source]#

convert parameters gradients of a module to a vector

Parameters
  • module (Module) -- module to convert the parameters of

  • clone (bool, optional) -- clones the output. Defaults to True.

  • detach (bool, optional) -- detaches the output. Defaults to True.

Returns

Module --

1-D Tensor of gradients of all parameters in the module. None if at

least grad of one children deos not exist.