当图例只是数值时，如何使用pyplot.scatter（）绘制图例？

Question

I cannot see to find (or figure out) how to show the legend using pyplot.scatter() when I've changed the labels to be numerical values.

That is, I transformed my categorical values 'a', 'b', 'c', .. into 0, 1, 2, ...

http://matplotlib.org/api/pyplot_api.html#matplotlib.pyplot.scatter

Here is the example given on the URL above:

import numpy as np
import matplotlib.pyplot as plt

N = 50
x = np.random.rand(N)
y = np.random.rand(N)
colors = np.random.rand(N)
area = np.pi * (15 * np.random.rand(N))**2  # 0 to 15 point radii

plt.scatter(x, y, s=area, c=colors, alpha=0.5, cmap=cm.jet)
plt.show()

which outputs

Normally, I believe one would do something like this:

example = plt.scatter(x, y, s=area, c=colors, alpha=0.5, cmap=cm.jet)
plt.legend(handles=[example])
plt.show()

This does not output a legend corresponding each color in the plot with the number in the array.

/opt/local/Library/Frameworks/Python.framework/Versions/3.4/lib/python3.4/site-packages/matplotlib/axes/_axes.py:518: UserWarning: The handle <matplotlib.collections.PathCollection object at 0x1167d03c8> has a label of '_collection0' which cannot be automatically added to the legend.
  'legend.'.format(handle, label))

How can I output a legend which shows which number corresponds with the array value?

Answer 1

A legend won't work for your scatter plot because the scatter plot creates a single object and would only appear as a single item in a legend . Since the color of each point depends upon the colormap of the axes, you'll instead need to use a colorbar

plt.colorbar(example)

If you want a legend instead, you need to create a separate scatter plot for each group and then create a legend from that . An example of that is shown here

Answer 2

This is what I personally use for all of my matplotlib legends. I usually store my data in dictionaries and pd.Series objects.

def get_legend_markers(D_label_color, marker="o", marker_kws={"linestyle":""}):
    """
    Usage: plt.legend(*legend_vars(D_taxon_color),
                      loc="lower center",
                      bbox_to_anchor=(0.5,-0.15),
                      fancybox=True, shadow=True,
                      prop={'size':15})

    Input: Dictionary object of {label:color}
    Output: Tuple of markers and labels
    """
    markers = [plt.Line2D([0,0],[0,0],color=color, marker=marker, **marker_kws) for color in D_label_color.values()]
    return (markers, D_label_color.keys())

So you could something like:

D_label_color = {"A":"green", "B":"blue", "C":"red"}
fig, ax = plt.subplots()
x = np.linspace(0, 2*np.pi)
y = np.sin(x)
c = ["green"]*20 + ["blue"]*20 + ["red"]*10
ax.scatter(x=x, y=y, c=c)
ax.legend(*get_legend_markers(D_label_color)

If you don't know exactly which color is assigned to which point you can assign them with seaborn using sns.color_palette(n_colors) for each category and then make a list/vector of colors ( like c above ) and then give ax.scatter that color assignment. Hope this helps. If you're dealing with continuous data then i would go with a color bar like @suever mentioned above. You can't really have a legend with continuous data.