I think it's a good point about the colors not being distinctive, especially since some folks have mentioned that the visuals should be colorblind friendly -- we could pick colors from here: http://colorbrewer2.org/

about sub-plots, not sure but open to trialing

what do you think @ml4713 ?

I don't think changing colors helps here since we only have a very simple and limited data to visualize and it makes sense for RBF, poly and linear to pick largely overlapped support vectors. As long as those vectors are on top of each other, it will look a little messier.

Something like this maybe?

fig, (ax1, ax2, ax3) = plt.subplots(1, 3, sharey=True)
ax1.plot(X, y_rbf, color='m', lw=lw, label='RBF model')
ax1.scatter(X[svr_rbf.support_], y[svr_rbf.support_], facecolor="none",
            edgecolor="m", marker='8', label='rbf support vectors', s=50)
ax1.legend(loc='upper center', bbox_to_anchor=(0.5, 1.05),
          ncol=3, fancybox=True, shadow=True)

ax2.plot(X, y_lin, color='c', lw=lw, label='Linear model')
ax2.scatter(X[svr_lin.support_], y[svr_lin.support_], facecolor="none",
            edgecolor="c", marker='^', label='linear support vectors', s=50)
ax2.legend(loc='upper center', bbox_to_anchor=(0.5, 1.05),
          ncol=3, fancybox=True, shadow=True)

ax3.plot(X, y_poly, color='g', lw=lw, label='Polynomial model')
ax3.scatter(X[svr_poly.support_], y[svr_poly.support_], facecolor="none",
            edgecolor="g", marker='s', label='poly support vectors', s=50)
ax3.legend(loc='upper center', bbox_to_anchor=(0.5, 1.05),
          ncol=3, fancybox=True, shadow=True)

fig.text(0.5, 0.04, 'data', ha='center', va='center')
fig.text(0.06, 0.5, 'target', ha='center', va='center', rotation='vertical')
fig.suptitle("Support Vector Regression", fontsize=14)
plt.show()

hm I can't see the rendered docs?

strange, i am able to see the plot.
@vasudhathinks are you able to see?

yes, I can see the plots similar to rita's screenshot. I had to define lw = 1

Sorry I meant I can't find the artifact generated by circleCI, I haven't tried running locally.

Separating the plots is great. +1
I suggest to change the figure size (something like plt.subplots(1, 3, figsize=(10, 4))) to have something wider.
We also need to plot all the training points in each subplot (maybe with a black dot?).

@ml4713 Checking in to see how this PR is going.

@reshamas I just realized that the travis CI build failed and not sure why. other than this, i think I am done.

@ml4713
I looked through the travis test file and the error is:
Failed: The splits for data, are same even when random state is not set
and
/home/travis/build/scikit-learn/scikit-learn/sklearn/model_selection/tests/test_split.py:487: Failed

But, I don't know what they mean.
@amueller @jnothman Are you able to explain the error and how to fix it?

I don't know that failure, but we no longer test python 2.7 in master, so merging in updated master should fix it