使用np.vstack()时遇OSError: [WinError 87]参数错误的问题求助

问题详情 #

• 使用CPU多进程加速渲染32768×32768尺寸的曼德博集合，执行np.vstack(results)拼接多进程计算结果时，触发OSError: [WinError 87] The parameter is incorrect错误；16384×16384及以下尺寸可正常运行。
• 改用交互渲染方案后，最初存在显示异常，通过在update_plot函数中添加ax.cla()解决，但仍不清楚大尺寸拼接错误的根源。

相关代码 #

多进程大尺寸渲染代码 #

import numpy as np
import matplotlib.pyplot as plt
import multiprocessing as mp
import warnings
import time

start = time.time() #to time program

warnings.filterwarnings("ignore") #For the overflow and invalid exponent runtime warnings

b = 8192*4 #Length and Width of the image

if __name__ == "__main__":
    num_cores = mp.cpu_count() 
    Y, X = np.mgrid[-1 * (b // 2):b // 2, -1 * (b // 2):b // 2]  #include all four quadrants about (0,0), not just Quadrant I
    
    chunk_size = b // num_cores  
    chunks = []
    
    for i in range(num_cores):   #create the chunks
        start_row = i * chunk_size
        end_row = (i + 1) * chunk_size if i < num_cores - 1 else b
        chunks.append((Y, X, start_row, end_row))

    pool = mp.Pool(processes=num_cores)
    results = pool.imap(calculate_mandelbrot_chunk, chunks, 3) #this is the error line for b > 2^14
    pool.close()
    pool.join()
####vvvvvvvvvvvvvvvvvvvvvv##################
    a = np.vstack(results)  

交互渲染代码 #

import matplotlib.pyplot as plt
from mpl_interactions import ioff, panhandler, zoom_factory
import warnings
import time
import numpy as np

warnings.filterwarnings("ignore")  # For the overflow and invalid exponent runtime warnings

start = time.time()  # to time program

b = 512  # Length and Width of the image
xlims = (-512,512)
ylims = (-512,512)


#scroll code
def on_ylims_change(event_ax):
    global ylims
    ylims = event_ax.get_ylim()

def on_xlims_change(event_ax):
    global xlims
    xlims = event_ax.get_xlim()
    print("Updated xlims:", xlims)
    update_plot(xlims, ylims)


#calc mandelbrot set
def calculate_mandelbrot(xmin, xmax, ymin, ymax, size=512):
    Y, X = np.mgrid[ymin:ymax, xmin:xmax]
    c = (X + Y * 1j) * (4 / size)  # Complex numbers array
    c = c.astype(np.complex64)  # Save memory
    a = np.copy(c) * 0
    for k in range(64):
        a = a ** 2 + c
    a[a < 100] = 1  # Filter divergents out
    return np.real(a)  # Type casting

def update_plot(xlims, ylims):
    ax.cla()  # Added this line to fix display issues
    a_zoomed = calculate_mandelbrot(xlims[0], xlims[1], ylims[0], ylims[1], size=b)
    ax.imshow(a_zoomed, cmap="gray", extent=(xlims[0], xlims[1], ylims[0], ylims[1]))

def blank(xlims, ylims):
    a_zoomed = np.zeros([32, 32])
    ax.imshow(a_zoomed, cmap="gray", extent=(xlims[0], xlims[1], ylims[0], ylims[1]))

a = calculate_mandelbrot(xlims[0], xlims[1], ylims[0], ylims[1], size=b)

end = time.time()
print(end - start)

#### scroll wheel zoom ####
with plt.ioff():
    fig, ax = plt.subplots()
ax.imshow(a, cmap="gray", extent=(-512,512,-512,512))
disconnect_zoom = zoom_factory(ax)
ax.callbacks.connect('ylim_changed', on_ylims_change)
ax.callbacks.connect('xlim_changed', on_xlims_change)

pan_handler = panhandler(fig)

plt.show()

注：在Intel i9设备上，b=512时每次重绘仅需0.3055903911590576 seconds，添加ax.cla()后解决了交互渲染时的图像叠加显示问题。

内容的提问来源于stack exchange，提问作者Aarush Abrol