Re: [Tutor] how to speed up this code?

[Pujo Aji]

Hello Andrew,

 

Thanks for the info.

I wonder if everytime we have to solve loop inside loop which use big array will force us to think another algorithm like matrix model. Since this model is suitable with Numeric.py 

 

I think looping style sometimes easier to describe our problem.
 

pujo
 

On 10/13/05, Andrew P <[EMAIL PROTECTED]> wrote:

It's way past my bedtime, but any use of numpy/numarray that involves two nested for loops to step over each element is the wrong solution :)  You need to figure out how to get rid of that inner for.  That is what is slowing you down. 

Compare these two ways to multiply a 1000 element array by 100.  The first one steps over the elements one at a time, multiplying each one in turn.  The second multiplies the entire array at once.  Which boils down to looping over 2000 rows, instead of 4,000,000 elements :)

If I was more awake, I'd try to figure out how you can do that.  But this should give you an idea of what arrays are useful for, and how to approach the problem.

>>> def time_loop(num):
...     a = arange(1000)
...     b = zeros(1000)
...     t = time.clock()
...     for i in range(num):
...         for i in range(len(a)):
...             b[i] = a[i] * 100.0
...     print time.clock() - t
...    
>>> time_loop(100000)
59.7517100637


>>> def time_numeric(num):
...     a = arange(1000)
...     b = zeros(1000)
...     t = time.clock()
...     for i in range(num):
...         b = a*100
...     print time.clock() - t
...    
>>> time_numeric(100000)
1.44588097091

On 10/13/05, Pujo Aji < [EMAIL PROTECTED]> wrote:

I have code like this:

def f(x,y):
    return math.sin(x*y) + 8 * x
 

def main():
    n = 2000
    a = zeros((n,n), Float)
    xcoor = arange(0,1,1/float(n))
    ycoor = arange(0,1,1/float(n))

    for i in range(n):
        for j in range(n):
            a[i,j] = f(xcoor[i], ycoor[j])  # f(x,y) = sin(x*y) + 8*x

    print a[1000,1000]
    pass

if __name__ == '__main__':
    main()

I try to make this run faster even using psyco, but I found this still
slow, I tried using java and found it around 13x faster...

public class s1 {

        /**
         * @param args
         */
        public static int n = 2000;
        public static double[][] a = new double[n][n];
        public static double [] xcoor = new double[n];
        public static double [] ycoor = new double[n];

        public static void main(String[] args) {
                // TODO Auto-generated method stub
                for (int i=0; i<n; i++){
                        xcoor[i] = i/(float)(n);
                        ycoor[i] = i/(float)n;
                }

                for (int i=0; i<n; i++){
                        for (int j=0; j<n; j++){
                                a[i][j] = f(xcoor[i], ycoor[j]);
                        }
                }

                System.out.println(a[1000][1000]);

        }
        public static double f(double x, double y){
                return Math.sin(x*y) + 8*x;
        }

}

 

Can anybody help?

pujo

_______________________________________________
Tutor maillist  -  Tutor@python.org
http://mail.python.org/mailman/listinfo/tutor

_______________________________________________
Tutor maillist  -  Tutor@python.org
http://mail.python.org/mailman/listinfo/tutor