61 lines
1.9 KiB
Python
61 lines
1.9 KiB
Python
from typing import Any, Dict
|
|
import torch
|
|
import ast
|
|
|
|
|
|
class PtPermute:
|
|
"""
|
|
Permutes the dimensions of a PyTorch tensor according to the specified order.
|
|
For example, if a tensor has shape (2, 3, 96, 32) and `new_axes="[0, 3, 1, 2]"`,
|
|
the output will have shape (2, 32, 3, 96).
|
|
|
|
Specify the new axis order in the `new_axes` field using brackets, as shown below:
|
|
```
|
|
[0, 3, 1, 2]
|
|
```
|
|
category: PyTorch wrapper - Transform
|
|
"""
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls) -> Dict[str, Any]:
|
|
"""
|
|
Defines the input types for the function.
|
|
|
|
Returns:
|
|
Dict[str, Any]: A dictionary specifying required input types.
|
|
"""
|
|
return {
|
|
"required": {
|
|
"tens": ("TENSOR", {}),
|
|
"new_axes": ("STRING", {"multiline": True})
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES: tuple = ("TENSOR",)
|
|
FUNCTION: str = "f"
|
|
CATEGORY: str = "Data Analysis"
|
|
|
|
def f(self, tens: torch.Tensor, new_axes: str) -> tuple:
|
|
"""
|
|
Permutes the dimensions of the input tensor based on the specified order.
|
|
|
|
Args:
|
|
tens (torch.Tensor): The input tensor.
|
|
new_axes (str): A string representation of the desired dimension order,
|
|
e.g., "[0, 3, 1, 2]" or "(0, 3, 1, 2)".
|
|
|
|
Returns:
|
|
tuple: A tuple containing the permuted PyTorch tensor.
|
|
"""
|
|
# Convert string to a tuple of integers
|
|
permute_order = tuple(ast.literal_eval(new_axes))
|
|
|
|
# Ensure the parsed order is valid
|
|
if not isinstance(permute_order, tuple) or not all(isinstance(i, int) for i in permute_order):
|
|
raise ValueError(f"Invalid format for new_axes: {new_axes}. Must be a tuple or list of integers.")
|
|
|
|
# Apply permutation
|
|
tens = tens.permute(*permute_order)
|
|
|
|
return (tens,)
|