param_dict

snippets.param_dict.from_param_dict(params: ParamDict, shapes: Dict[str, Tuple[int]]) → TensorLike

Convert a dictionary of parameters to a tensor of parameters for batch processing.

Parameters:

• params – Dictionary mapping parameter names to tensors.

• shapes – Dictionary mapping parameter names to shapes.

Returns:

Tensor of concatenated, raveled parameters ordered by parameter name.

Example

>>> import torch
>>> from snippets.param_dict import from_param_dict

>>> params = {"b": torch.arange(12).reshape((3, 4)), "a": torch.arange(3)}
>>> shapes = {"b": (3, 4), "a": (3,)}
>>> from_param_dict(params, shapes)
tensor([ 0,  1,  2,  0, ..., 11])

snippets.param_dict.to_param_dict(params: TensorLike, shapes: Dict[str, Tuple[int]]) → ParamDict

Convert a tensor of parameters to a dictionary of parameters.

Parameters:

• params – Tensor of concatenated, raveled parameters ordered by parameter name.

• shapes – Dictionary mapping parameter names to shapes.

Returns:

Dictionary mapping parameter names to tensors with the specified shape.

Example

>>> import torch
>>> from snippets.param_dict import to_param_dict

>>> params = torch.arange(15)
>>> shapes = {"b": (3, 4), "a": (3,)}
>>> to_param_dict(params, shapes)  
{'a': tensor([0, 1, 2]), 'b': tensor([[ 3,  4,  5,  6], ...])}