Question

假设我有一个形状为n X m的二维numpy数组（其中n是大数且m> = 1）。每列代表一个属性。下面提供了n = 5，m = 3的示例：

[[1,2,3],
[4,5,6],
[7,8,9],
[10,11,12],
[13,14,15]]

我想用history_steps = p（1＆lt; p＆lt; = n）训练我的模型在属性的历史上。对于p = 2，我预期的输出（形状（n-p + 1×m * p））是

[[1,4,2,5,3,6],
[4,7,5,8,6,9],
[7,10,8,11,9,12],
[10,13,11,14,12,15]]

我尝试通过分离列然后连接输出来在pandas中实现它。

def buff(s, n):
    return (pd.concat([s.shift(-i) for i in range(n)], axis=1).dropna().astype(float))

但是，就我的目的而言，基于numpy的方法会更好。另外，我想避免分裂和连接。

我该怎么做？

Answer 1

这是一种基于NumPy的方法，使用np.lib.stride_tricks.as_strided -

专注于性能

    @FXML

public void generateButton(ActionEvent event) {

    String fingerprint = fingerprintText.getText().toLowerCase();
    String erg = Verifier.getDdProUnlockPIN(fingerprint);
    pinField.setText(erg);
    copyText = erg;
    log.info("Pin " + erg + "wird generiert");
}

示例运行 -

def strided_axis0(a, L = 2):
    # INPUTS :
    # a : Input array
    # L : Length along rows to be cut to create per subarray

    # Store shape and strides info
    m,n = a.shape
    s0,s1 = a.strides
    nrows = m - L + 1

    strided = np.lib.stride_tricks.as_strided

    # Finally use strides to get the 3D array view and then reshape
    return strided(a, shape=(nrows,n,L), strides=(s0,s1,s0)).reshape(nrows,-1)

Answer 2

您可以使用dstack + reshape：

a = np.array([[1,2,3],
[4,5,6],
[7,8,9],
[10,11,12],
[13,14,15]])

# use `dstack` to stack the two arrays(one with last row removed, the other with first 
# row removed), along the third axis, and then use reshape to flatten the second and third
# dimensions
np.dstack([a[:-1], a[1:]]).reshape(a.shape[0]-1, -1)

#array([[ 1,  4,  2,  5,  3,  6],
#       [ 4,  7,  5,  8,  6,  9],
#       [ 7, 10,  8, 11,  9, 12],
#       [10, 13, 11, 14, 12, 15]])

要概括为任意p，请使用列表推导生成已移位数组的列表，然后执行stack+reshape：

n, m = a.shape
p = 3
np.dstack([a[i:(n-p+i+1)] for i in range(p)]).reshape(n-p+1, -1)

#array([[ 1,  4,  7,  2,  5,  8,  3,  6,  9],
#       [ 4,  7, 10,  5,  8, 11,  6,  9, 12],
#       [ 7, 10, 13,  8, 11, 14,  9, 12, 15]])

从2d numpy数组创建数据历史记录？

2 个答案: