解决嵌套列表转NumPy数组形状问题：将(a,1,c)转(a,)

1. 单个(a,1,c)数组转(a,)的方法 #

手动遍历赋值 #

直接创建空的object类型数组，将(a,1,c)数组中每个位置的[0,:]子数组赋值进去：

import numpy as np

# 示例：形状为(2,1,3)的数组
arr = np.array([[[1,2,3]], [[5,4,7]]])
# 创建形状为(2,)的object数组
converted_arr = np.empty(arr.shape[0], dtype=object)
for i in range(arr.shape[0]):
    converted_arr[i] = arr[i, 0, :]

print(converted_arr.shape)  # 输出 (2,)
print(converted_arr[0])     # 输出 [1 2 3]

列表推导式简化 #

用列表推导式提取子元素后，再转为object数组：

arr = np.array([[[1,2,3]], [[5,4,7]]])
converted_arr = np.array([sub[0] for sub in arr], dtype=object)

print(converted_arr.shape)  # 输出 (2,)

2. 整合到outer_list的完整处理流程 #

针对你的嵌套列表，先统一处理所有内部数组，确保它们都转为(a,)形状的object数组，再组合成外层数组并保存：

import numpy as np

outer_list = [
    [
        [[1, 2, 3], [4, 6, 7]],
        [[7, 8, 9], [11, 15, 17]]
    ],
    [
        [[5, 6, 7], [3, 4, 6], [1, 2, 4]],
        [[1, 6, 3], [3, 6, 4], [34, 2, 1]]
    ],
    [
        [[1, 2, 3]],
        [[5, 4, 7]]
    ]
]

processed_outer = []
for inner_list in outer_list:
    processed_inner = []
    for sub_list in inner_list:
        arr = np.array(sub_list)
        # 判断是否为需要转换的(a,1,c)形状
        if arr.ndim == 3 and arr.shape[1] == 1:
            converted = np.array([x[0] for x in arr], dtype=object)
        else:
            converted = np.array(sub_list, dtype=object)
        processed_inner.append(converted)
    processed_outer.append(np.array(processed_inner, dtype=object))

# 生成最终可保存的数组
final_arr = np.array(processed_outer, dtype=object)
# 保存为.npy文件
np.save("output.npy", final_arr)

内容的提问来源于stack exchange，提问作者Mistakamikaze