These options do not work...
import numpy as np
import matplotlib.pyplot as plt
arr = np.random.random((5,3))
ax = plt.axes()
ax.scatter(arr[:,0],arr[:,1],
It seems like you are trying to populate the legend with the actual scatter plot, or at least reference what is going on in the scatter plot. To create a legend, you need to draw it as a separate entity - meaning that the scatter point shapes and colors need to be recreated, for example as a subplot. This is a slightly more manual approach but should work:
colors = ['k','r','g','r','b']
ax = plt.axes()
ax.scatter(arr[:,0],arr[:,1],c=['k','r','g','r','b'])
line1 = plt.Line2D(range(10), range(10), marker='o', color=colors[0])
line2 = plt.Line2D(range(10), range(10), marker='o',color=colors[1])
line3 = plt.Line2D(range(10), range(10), marker='o',color=colors[2])
line4 = plt.Line2D(range(10), range(10), marker='o',color=colors[3])
line5 = plt.Line2D(range(10), range(10), marker='o',color=colors[4])
plt.legend((line1,line2,line3, line4, line5),('color1','color2', 'color3', 'color4', 'color5'),numpoints=1, loc=1)
plt.show()