ValueError: setting an array element with a sequence...

报错：ValueError: setting an array element with a sequence…

案例1：numpy库使用numpy.array转换list

a是一个list，其包含两个元组，使用np.array进行转换，会报错：

import numpy as np
a = ([([1, 2, 3], 0, 1.0, [3, 4], 0), ([1, 2, 3], 0, 1.0, [3, 4], 0)])
a = np.array(a)

ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 2 dimensions. The detected shape was (2, 5) + inhomogeneous part.

有的numpy版本不报错而是警告：

VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify ‘dtype=object’ when creating the ndarray.

加上 dtype=object 就不会再报错或警告：

import numpy as np
a = ([([1, 2, 3], 0, 1.0, [3, 4], 0), ([1, 2, 3], 0, 1.0, [3, 4], 0)])
a = np.array(a, dtype=object)

---

简单介绍一下此处的背景，在python中，list无法被数组进行索引：

a = ([([1, 2, 3], 0, 1.0, [3, 4], 0), ([1, 2, 3], 0, 1.0, [3, 4], 0)])
index = [1, 0]
b = a[index]
print(f'{b}')

TypeError: list indices must be integers or slices, not list

将list转换成numpy，然后就可以使用数组进行索引了：

import numpy as np
a = ([([1, 2, 3], 0, 1.0, [3, 4], 0), ([1, 2, 3], 0, 1.0, [3, 4], 0)])
a = np.array(a, dtype=object)
index = [1, 0]
b = a[index]
print(f'{b}')

输出为：

案例2：新版gym库使用state = env.reset()

新版本gym库的env.reset()返回是两个元组，其中第二个为dict，一般不包含数据。若直接使用state = env.reset()返回的数值如下所示：

因此，继续原来的操作（状态分析转换等等）就会报错ValueError: setting an array element with a sequence，正确的写法如下所示，这样就能直接得到agent所对应的状态了：

state, _ = env.reset()

同理，env.step(action)的返回值也有所改变，返回参数由原来的四个改为了五个，如果使用下面代码则会报错（ValueError: too many values to unpack (expected 4)）：

next_state, reward, done, _ = env.step(action)

需要使用如下的修改代码：

next_state, reward, done, _, _ = env.step(action)

新版gym库返回参数的详细定义可以参考官方网站：
https://www.gymlibrary.dev/api/core/