python - 在python中有效读取没有分隔符的数字矩阵

Question

我有一个文件，其中包含数字 [0-9] 矩阵，不带分隔符，形状为 (N, M)。 N 约为 50k，M 约为 50k。 例如，矩阵文件的小版本是 mat.txt

0012230012000
0012230002300
0012230004200

现在我正在使用以下代码，但我对速度不太满意。

def read_int_mat(path):
    """
    Read a matrix of integer with [0-9], and with no delimiter.
    """
    with open(path) as f:
        mat = np.array(
            [np.array([int(c) for c in line.strip()]) for line in f.readlines()],
            dtype=np.int8,
        )
    return mat

编辑: 这是一个迷你基准

import numpy as np
def read_int_mat(path):
    """
    Read a matrix of integer with [0-9], and with no delimiter.
    """
    with open(path) as f:
        mat = np.array(
            [np.array([int(c) for c in line.strip()]) for line in f.readlines()],
            dtype=np.int8,
        )
    return mat

%timeit read_int_mat("mat.txt")
%timeit np.genfromtxt("mat.txt", delimiter=1, dtype="int8")

print(read_int_mat("mat.txt"))
print(np.genfromtxt("mat.txt", delimiter=1, dtype="int8"))

输出是:

61.6 µs ± 1.32 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)
327 µs ± 4.95 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
[[0 0 1 2 2 3 0 0 1 2 0 0 0]
 [0 0 1 2 2 3 0 0 0 2 3 0 0]
 [0 0 1 2 2 3 0 0 0 4 2 0 0]]
[[0 0 1 2 2 3 0 0 1 2 0 0 0]
 [0 0 1 2 2 3 0 0 0 2 3 0 0]
 [0 0 1 2 2 3 0 0 0 4 2 0 0]]

有什么我可以尝试进一步加快速度的吗？ Cython 会帮忙吗？非常感谢。

Answer 1

您可以使用np.genfromtxt ，例如:

文件(13 列):

0012230012000
0012230002300
0012230004200

然后:

x = np.genfromtxt("file.txt", delimiter=1, dtype="int8")
print(x)

打印:

[[0 0 1 2 2 3 0 0 1 2 0 0 0]
 [0 0 1 2 2 3 0 0 0 2 3 0 0]
 [0 0 1 2 2 3 0 0 0 4 2 0 0]]

---

编辑:带有 np.fromiter 的版本并以二进制模式打开文件:

def read_npfromiter(path):
    with open(path, "rb") as f:
        return np.array(
            [np.fromiter((chr(c) for c in l.strip()), dtype="int8") for l in f],
        )

---

形状为(168, 9360)的文件基准:

from timeit import timeit


def read_int_mat(path):
    """
    Read a matrix of integer with [0-9], and with no delimiter.
    """
    with open(path, "r") as f:
        mat = np.array(
            [
                np.array([int(c) for c in line.strip()])
                for line in f.readlines()
            ],
            dtype=np.int8,
        )
    return mat


def read_npfromiter(path):
    with open(path, "rb") as f:
        return np.array(
            [np.fromiter((chr(c) for c in l.strip()), dtype="int8") for l in f],
        )


def f1(f):
    return np.genfromtxt(
        f, delimiter=1, dtype="int8", autostrip=False, encoding="ascii"
    )


def f2(f):
    return read_int_mat(f)


def f3(f):
    return read_npfromiter(f)


t1 = timeit(lambda: f1("file.txt"), number=1)
t2 = timeit(lambda: f2("file.txt"), number=1)
t3 = timeit(lambda: f3("file.txt"), number=1)

print(t1)
print(t2)
print(t3)

结果:

1.0680423599551432
0.28135157003998756
0.19099885696778074