1 files changed, 42 insertions, 0 deletions
diff --git a/benchmarks/simpleindex.py b/benchmarks/simpleindex.py
new file mode 100644
index 000000000..6b1a63d34
--- /dev/null
+++ b/benchmarks/simpleindex.py
@@ -0,0 +1,42 @@
+import timeit
+# This is to show that NumPy is a poorer choice than nested Python lists
+#   if you are writing nested for loops.
+# This is slower than Numeric was but Numeric was slower than Python lists were
+#   in the first place. 
+
+N = 30
+code2 = r"""
+for k in xrange(%d):
+    for l in xrange(%d):
+        res = a[k,l].item() + a[l,k].item()
+""" % (N,N)
+code3 = r"""
+for k in xrange(%d):
+    for l in xrange(%d):
+        res = a[k][l] + a[l][k]
+""" % (N,N)
+code = r"""
+for k in xrange(%d):
+    for l in xrange(%d):
+        res = a[k,l] + a[l,k]
+""" % (N,N)
+setup3 = r"""
+import random
+a = [[None for k in xrange(%d)] for l in xrange(%d)]
+for k in xrange(%d):
+    for l in xrange(%d):
+        a[k][l] = random.random()
+""" % (N,N,N,N)
+t1 = timeit.Timer(code, 'import numpy as N; a = N.random.rand(%d,%d)' % (N,N))
+t2 = timeit.Timer(code, 'import MLab as N; a=N.rand(%d,%d)' % (N,N))
+t3 = timeit.Timer(code, 'import numarray.mlab as N; a=N.rand(%d,%d)' % (N,N))
+t4 = timeit.Timer(code2, 'import numpy as N; a = N.random.rand(%d,%d)' % (N,N))
+t5 = timeit.Timer(code3, setup3)
+t6 = timeit.Timer("res = a + a.transpose()","import numpy as N; a=N.random.rand(%d,%d)" % (N,N))
+print "shape = ", (N,N)
+print "NumPy 1: ", t1.repeat(3,100)
+print "NumPy 2: ", t4.repeat(3,100)
+print "Numeric: ", t2.repeat(3,100)
+print "Numarray: ", t3.repeat(3,100)
+print "Python: ", t5.repeat(3,100)
+print "Optimized: ", t6.repeat(3,100)