如何从数组中删除所有字符，只留下数字？

Question

我正在使用 sklearn.MLPClassifer 并将我的权重导出到一个 txt 文件中：

arr = list(np.array(model.coefs_, dtype = object))
file = open("weightss.txt", "w+")
file.write (str(arr))
file.close()

事实是，model.coefs_ 中每个 [] 的权重都不相同，并且它导出为“数组”和一堆我不知道的其他字符想要在里面。如何从此数组中仅导出数字（浮点类型）？

我得到的输出：

array([[ 0.055124  ,  0.04365641, -0.0271139 , ...,  0.02899075,
         -0.14442606, -0.04212195],
        [-0.01849607, -0.01092366,  0.01105672, ..., -0.00383433,
         -0.23810516,  0.03760549],
        [ 0.02357766, -0.04846305,  0.0028944 , ..., -0.00372297,
         -0.23913875,  0.03809774],
        ...,
        [ 0.04734891,  0.00075733,  0.05273402, ...,  0.02994463,
         -0.00568819, -0.05830916],
        [ 0.12168126,  0.06481046,  0.03357674, ..., -0.06854297,
          0.00458801, -0.06863405],
        [ 0.09052689, -0.03500097, -0.06934987, ..., -0.05713005,
         -0.04042818, -0.07102473]])
 array([[ 6.49345111e-01, -2.80867026e+00, -4.47025490e-01,
          5.47926183e-02, -2.32258820e-01,  3.00891945e-01,
         -3.28820315e+00, -1.11907300e+00,  2.13128839e-01,
         -2.60551663e+00,  2.42408007e+00, -1.31142015e+00,
         -7.13036636e-01,  2.98367056e+00,  1.23166718e+00,
          1.84157657e+00,  2.50413248e-01, -6.74166192e-01,
         -1.04169355e-01, -8.85277883e-01],
        [ 3.82943554e-02, -1.3093057

等等...... 我希望它只是数字，没有那些“，”和其他字符，所以我可以将其导入回来以使其学习这些权重。

Answer 1

您可以尝试以下操作：

import os

with open('weightss.txt', 'w') as fp:
    for a in model.coefs_:
        np.savetxt(fp, a, fmt='%f')
        fp.write(os.linesep)

演示：

rng = np.random.default_rng(2022)
coefs = [np.array(rng.random((4, 5))), np.array(rng.random((5, 1)))]
print(coefs)

# Output
[array([[0.24742606, 0.09299006, 0.61176337, 0.06066207, 0.66103343],
        [0.75515778, 0.1108689 , 0.04305584, 0.41441747, 0.98862926],
        [0.96919869, 0.25697153, 0.55876211, 0.24234798, 0.32202029],
        [0.89135975, 0.94611366, 0.72253931, 0.92847437, 0.99608701]]),
 array([[0.2494223 ],
        [0.06229007],
        [0.94479027],
        [0.65028587],
        [0.32167568]])]

weightss.txt 的内容：

0.247426 0.092990 0.611763 0.060662 0.661033
0.755158 0.110869 0.043056 0.414417 0.988629
0.969199 0.256972 0.558762 0.242348 0.322020
0.891360 0.946114 0.722539 0.928474 0.996087

0.249422
0.062290
0.944790
0.650286
0.321676