如何从Python numpy 3D数组到2D再到1D再到2D（保留3D数组的原始第二和第三维）

我有一个3D阵列，一个：

`print(a.shape)
In [1]:(4, 4571, 8893)
b = a.reshape(a.shape[2]*a.shape[1],a.shape[0]) # Here I've also tried changing the shape of with (a.shape[2]*a.shape[1],a.shape[0])
print(b.shape)
In [2]:(40649903, 4)
c=some_function(b) # returns c which has same shape as b.shape[0]
print(c.shape)
In [2]: (40649903,)
d = c.reshape(a.shape[1],a.shape[2]) # same shape as a.shape[1:]
print(d.shape)
In [3]:(4571, 8893)

`

现在，当我看d时，我得到了如下形状：

plt.imshow(d)

但它必须像下图所示（pl.忽略颜色，黄色区域的形状必须像海军蓝区域）：

---

也许这与重塑轴有关，但我无法确定在哪里使用了错误的轴来重塑。我对此做了一些思考，并阅读了numpy文档，但是文档和在线示例（所以问题）似乎没有针对我的具体问题的清晰示例。任何我错过的方向都会有帮助

以下是其他可能有同样问题的人的答案：

a = np.arange(24).reshape(4,3,2)
print(a); print(a.shape)

b = a.reshape(a.shape[0],a.shape[1]*a.shape[2]).T; 
print(b); print(b.shape) # X

c = a[0].flatten() # Y
print(c); print(c.shape)

d = c.reshape(a[1].shape); 
print(d); print(d.shape) # same as print(a[0].shape)

---

谢谢你的建议@hpaulj

尝试一些小东西，比如np.arange（24.重塑（4,3,2）。重塑为（4,6）与重塑为（6,4）完全不同。好主意，谢谢！我没想到！！

a = np.arange(24).reshape(4,3,2)
print(a); print(a.shape)

b = a.reshape(a.shape[0],a.shape[1]*a.shape[2]).T; 
print(b); print(b.shape) # X

c = a[0].flatten() # Y
print(c); print(c.shape)

d = c.reshape(a[1].shape); 
print(d); print(d.shape) # same as print(a[0].shape)