[에러랑 코드] AttributeError: 'tuple' object has no attribute 'shape' / 트리 구조로 만들어서 빠르게 파악하기

 

1. Tuple 여러개를 트리 구조로 보고 싶어졌다

 

 

AttributeError: 'tuple' object has no attribute 'shape'

 

tuple 구조일 때 shape을 보기 힘들어서 여러 dimension인 자료구조에 대해 대응하기 힘들었다 

 

 

 

 

 

한번 len으로 알더라도 중첩되면 계속 에러 나오는게 신경쓰였다.

 

 

 

 

 

 

2. 코드 

 

import torch

def inspect_shapes(structure, prefix=''):
    """
    Recursively inspect and print the shapes and lengths of tensors,
    tuples, or lists within a given structure.
    
    Args:
        structure: The data structure (tensor, tuple, or list) to inspect.
        prefix: The prefix for print statements to indicate nesting level.
    """
    if isinstance(structure, torch.Tensor):
        # If it's a tensor, print its shape
        print(f"{prefix}Tensor with shape: {structure.shape}")
    elif isinstance(structure, (list, tuple)):
        # If it's a list or tuple, print its length
        print(f"{prefix}{type(structure).__name__} with length: {len(structure)}")
        # Recursively inspect each element
        for i, item in enumerate(structure):
            inspect_shapes(item, prefix=prefix + f'  [item {i}] ')
    else:
        # Unexpected type
        print(f"{prefix}Unexpected type: {type(structure)}")

# Example usage within a hook or standalone
example_output = (
    torch.randn(2, 128, 64, 64),  # A tensor
    [torch.randn(2, 64, 32, 32), torch.randn(2, 32, 16, 16)],  # A list of tensors
    (torch.randn(2, 16, 8, 8),)  # A tuple containing a tensor
)

# Inspect the structure to check shapes and lengths
inspect_shapes(example_output)

 

 

 

 

여러 구조로 tuple, list가 섞인 경우 어떻게 생겼는지 트리 구조로 쓰도록 코드 구해서 잘 사용했다.