_ArrayMemoryError: 无法分配

Question

我正在尝试处理数据，但我经常遇到此错误：numpy.core._exceptions._ArrayMemoryError: Unable to allocate 8.00 GiB for an array with shape (32761, 32761) and data type float64 这是我的代码：

import numpy as np
import csv
import tkinter as tk
from tkinter import filedialog

"""
Importing data 
"""

root= tk.Tk()

canvas1 = tk.Canvas(root, width = 300, height = 300, bg = 'gray1', relief = 'raised')
canvas1.pack()

def getCSV ():
    global df
    
    import_file_path = filedialog.askopenfilename()
    df = np.genfromtxt(import_file_path, delimiter=' ')
    
    print (df)
    
browseButton_CSV = tk.Button(text="      Import CSV File     ", command=getCSV, bg='OrangeRed4', fg='black', font=('helvetica', 12, 'bold'))
canvas1.create_window(150, 150, window=browseButton_CSV)

root.mainloop()

x, y, mag = df[:,0], df[:,1], df[:,3]
Xshape, Yshape, MAGshape = np.shape(x), np.shape(y), np.shape(mag)


def fftfreqs(x, y, shape, windowLen):
    """
    Get two 2D-arrays with wavenumbers [rads/km] in x, y directions.
    """
    nx = ny = shape[0]
    
    dx = (x.max() - x.min())/(nx - 1)           # Spacing
    fx = 2*np.pi*np.fft.fftfreq(windowLen[0], dx)
    
    dy = (y.max() - y.min())/(ny - 1)           # Spacing
    fy = 2*np.pi*np.fft.fftfreq(windowLen[1], dy)
    
    return np.meshgrid(fy, fx)[::-1]


"""
Calculation of power spectrum density
"""
shap = (np.shape(x)[0], np.shape(y)[0])
kx, ky = fftfreqs(x, y, shap, shap)
pds = (abs(np.fft.fft2(np.reshape(mag, (1,shap[0])))))**2

"""
Calculation of Radially Averaged Power Spectrum
"""

nx, ny = pds.shape

max_radius = min(kx.max(), ky.max())

ring_width = max(np.unique(kx)[np.unique(kx) > 0][0], np.unique(ky)[np.unique(ky) > 0][0])

k = np.sqrt(kx**2 + ky**2)
pds_radial = []
k_radial = []
radius_i = -1
while True:
    radius_i += 1
    if radius_i*ring_width > max_radius:
        break
    else:
        if radius_i == 0:
            inside = k <= 0.5*ring_width
        else:
            inside = np.logical_and(k > (radius_i - 0.5)*ring_width, k <= (radius_i + 0.5)*ring_width)
        pds_radial.append(pds[inside].mean())
        k_radial.append(radius_i*ring_width)

我在 8GB RAM 系统上运行它，但我也尝试在 GOOGLE COLAB 上运行它，但结果相同。 提前致谢

Answer 1

我远不是快速傅立叶变换的专家，所以我无法告诉你你所做的是否有意义。但是，我认为您的 MemoryError 来自这样一个事实，即您在同一命令中处理的不仅仅是一个 (32761, 32761) 数组。所以，也许你可以分配第一个，但也许不能分配第二个；你明白了。看看我在下面的建议，如果这对您有所帮助，请告诉我。

import numpy as np
import tkinter as tk
from tkinter import filedialog


def getCSV():
    global df
    import_file_path = filedialog.askopenfilename()
    df = np.genfromtxt(import_file_path, delimiter=' ')


def fftfreqs():
    """
    Get two 2D-arrays with wavenumbers [rads/km] in x, y directions.
    """
    nx = ny = df.shape[0]

    dx = (x.max() - x.min()) / (nx - 1)  # Spacing
    fx = 2 * np.pi * np.fft.fftfreq(nx, dx)

    dy = (y.max() - y.min()) / (ny - 1)  # Spacing
    fy = 2 * np.pi * np.fft.fftfreq(ny, dy)

    return np.meshgrid(fy, fx)[::-1]


"""
Importing data
"""

root = tk.Tk()

canvas1 = tk.Canvas(root, width=300, height=300, bg='gray1', relief='raised')
canvas1.pack()

browseButton_CSV = tk.Button(text="      Import CSV File     ", command=getCSV,
                             bg='OrangeRed4', fg='black',
                             font=('helvetica', 12, 'bold'))
canvas1.create_window(150, 150, window=browseButton_CSV)

root.mainloop()

x, y, mag = df[:, 0], df[:, 1], df[:, 3]

"""
Calculation of power spectrum density
"""
kx, ky = fftfreqs()
pds = np.fft.fft2(np.reshape(mag, (1, df.shape[0]))) ** 2

"""
Calculation of Radially Averaged Power Spectrum
"""

max_radius = min(kx.max(), ky.max())
print(kx.shape)  # (32761, 32761)
print(ky.shape)  # (32761, 32761)
# Do not do this
ring_width = max(np.unique(kx)[np.unique(kx) > 0]
                 [0], np.unique(ky)[np.unique(ky) > 0][0])
# Do something like this instead
kx_unique = np.unique(kx)
# Process kx_unique
del kx_unique
ky_unique = np.unique(ky)
# Process ky_unique
del ky_unique
ring_width = max(, )

k = np.sqrt(kx**2 + ky**2)
pds_radial = []
k_radial = []
radius_i = -1
while True:
    radius_i += 1
    if radius_i * ring_width > max_radius:
        break
    else:
        if radius_i == 0:
            inside = k <= 0.5 * ring_width
        else:
            inside = np.logical_and(
                k > (radius_i - 0.5) * ring_width,
                k <= (radius_i + 0.5) * ring_width)
        pds_radial.append(pds[inside].mean())
        k_radial.append(radius_i * ring_width)