numpy.matrix.A1 和 ravel 的区别 [英] Difference between numpy.matrix.A1 and ravel

问题描述

据我所知，我还没有看到任何帖子对此进行评分，如果有的话，请注意我.

To my knowledge, I haven't seen any posts regrading this, if any, please note me.

根据 SciPy 网站，numpy.matrix.A1 等价于 np.asarray(x).ravel().

一个例子就足以说明问题:

One example will be enough to illustrate the problem:

x = np.matrix(np.arange(12).reshape((1, -1)))


print("Shape of x: ", x.shape)
print("Shape of x with asarray: ", np.asarray(x).shape)
print("Equality: ", np.array_equal(x, np.asarray(x)))
print("Shape of x ravel flatten: ", x.ravel().shape)
print("Shape of x ravel flatten with asarray: ", np.asarray(x).ravel().shape)

打印:

Shape of x:  (1, 12)
Shape of x with asarray:  (1, 12)
Equality:  True
Shape of x ravel flatten:  (1, 12)
Shape of x ravel flatten with asarray:  (12,)

问题:

正如观察到的，扁平化数组的维度与asarray不同，只是想知道为什么它会呈现这种维度不一致?

As observed, the dimension of flattened array is different with asarray, just wondered why it's being presenting such inconsistencies in dimensions?

从asarray函数的np实现来看，我没有看到任何可能导致维度问题的东西，而且它通过了相等测试(x ==np.asarray(x)).但除此之外，还有什么可能对数组进行隐式更改.

From np implementation of asarray function, I didn't see any thing might cause a dimension problem, plus it passes the equality test (x == np.asarray(x)). But other than this, what could be possibly making implicit changes to the array.

def asarray(a, dtype=None, order=None):
    return array(a, dtype, copy=False, order=order)

已

这可能会令人困惑

加上它通过了相等性测试 (x == np.asarray(x))

plus it passes the equality test (x == np.asarray(x))

更准确地说，我的意思是，它通过了相等性测试 (np.array_equal(x, np.asarray(x)))

to be more precise, I mean, it passes the equality test (np.array_equal(x, np.asarray(x)))

推荐答案

NumPy 矩阵 (np.matrix) 总是 2D.(数学上，严格定义矩阵，而不是矩阵或向量.)

NumPy matrices (np.matrix) are always 2D. (Mathematically, the strict definition of a matrix, rather than a matrix or vector.)

来自np.matrix.ravel:

返回展平后的矩阵 (1, N) 其中 N 是数字原始矩阵中的元素.

Return the matrix flattened to shape (1, N) where N is the number of elements in the original matrix.

NumPy 矩阵的一些动机是为了 Matlab 用户.请参阅此处了解一些NumPy matrix 与 array 上的更好点.

Some motivation for NumPy matrices is for Matlab users. See here for some of the finer points on NumPy matrix versus array.

简而言之，一个 NumPy 数组(这里是 asarray(x) 的结果)可以是一维结构.矩阵最小可以是 2d.type(np.asarray(x)) 毫不奇怪，是一个数组.(不要与 np.asanyarray()，你的结果将是一个矩阵，因为它是一个数组子类.

In brief, a NumPy array (the result of asarray(x) here) can be a 1-dimensional structure. Matrices can be a minimum of 2d. type(np.asarray(x)) is, not shockingly, an array. (Not to be confused with np.asanyarray(), for which your result would be a matrix because it's an array subclass.

最后，您注意到:

它通过了相等性测试 (x == np.asarray(x))

it passes the equality test (x == np.asarray(x))

我明白这可能有点令人困惑.从技术上讲，您希望使用 np.array_equal(x, np.asarray(x))，尽管它的计算结果仍为 True.然而，NumPy 逻辑测试通常意味着数据结构不可知，一般来说:

I see how this could be a bit confusing. Technically, you want to use np.array_equal(x, np.asarray(x)), although that still evaluates to True. However, NumPy logic testing is generally meant to be data-structure agnostic, in general:

np.array_equal([1, 2, 3], np.array([1, 2, 3]))
# True

(这是 来源code>array_equal()——两者都被转换为数组.)

(Here is the source for array_equal()--both are cast to arrays.)

底线是它们的最小维度"不同，一个是另一个的子类.

The bottom line is that their "minimum dimensionalities" are different, and one is a subclass of the other.

issubclass(np.matrix, np.ndarray)
# True

这篇关于numpy.matrix.A1 和 ravel 的区别的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！