Parallel Python to make fractal and resource serialized under pp

Question

I make one python script to make fractal image.

I try to use pp python module to make faster this source code.

The big issue is: image.putpixel((x, y), (i % 8 * 16, i % 4 * 32,i % 2 * 64)). This line of source code when is use like job give me somethig about : cPickle.UnpickleableError: Cannot pickle objects.

I think this resource cannot be serialized under pp. Any idea ? Thank you. Regards.

My source code :

from PIL import Image
#size of image
imgx = 600
imgy = 400
#make image buffer
image = Image.new("RGB", (imgx, imgy))

# area of fractal
xa = -2.0
xb = 2.0
ya = -2.0
yb = 2.0

#define constants
max_iterations = 10 # max iterations allowed
step_derivat = 0.002e-1 # step size for numerical derivative
error = 5e-19 # max error allowed

# function will generate the newton fractal
def f(z): return z * z  +complex(-0.31,0.031)

# draw derivate fractal for each y and x 
for y in range(imgy):
 zy = y * (yb - ya)/(imgy - 1) + ya
 for x in range(imgx):
  zx = x * (xb - xa)/(imgx - 1) + xa
  z = complex(zx, zy)
  for i in range(max_iterations):
   # make complex numerical derivative
   dz = (f(z + complex(step_derivat, step_derivat)) - f(z))/complex(step_derivat,step_derivat)
    # Newton iteration see wikipedia   
   z0 = z - f(z)/dz 
   # stop to the error 
   if abs(z0 - z) < error: 
    break
   z = z0
  #I use modulo operation expression to do RGB colors of the pixels 
  image.putpixel((x, y), (i % 8 * 16, i % 4 * 32,i % 2 * 64))

#save the result 
image.save("fractal.png", "PNG")

Answer 1

Parallel Python or the multiprocessing module require the objects which need to be passed around to other processes to be picklable, and the image object from PIL is not.

I recommend removing the call to image.putpixel from the function to be parallelized, returnin a simple list of RGB points or a numpy array which are picklable. then, after the calculations are done you can assemble the image.

Also, in order to get more specific advice, you should post the parallel version of the code as well.