BUG: Fixes color selection in andrews_curve

[tacaswell]

Fixed bug introduced in b5265a5 which change from picking a single random color for the class to picking a random color for each line.

This returns the color selection to a per-set basis.

See http://stackoverflow.com/questions/19667209/colors-in-andrews-curves

[jtratner]

Please post an image of output you get from this.

[jtratner]

(and nice to have reproducible example, even if it's just copying from SO question)

[tacaswell]

import pandas as pd
import numpy as np

df = pd.DataFrame(np.random.randn(100,3), columns=['A','B', 'C'])
df['X'] = np.random.choice(['Alpha', 'Beta', 'Theta'], size=100)
pd.tools.plotting.andrews_curves(df, 'X')

[jtratner]

thanks for posting that - this looks good to me (going to try to pull down and see if I get the same thing) - anyone have any comments?

[amanahuja]

Looks great now; thanks!

Let me know if this needs to be a separate request, but consider:

plt.ylabel(r'$f_x(t)$')
plt.xlabel('t')
ax.xaxis.set_ticks([-pi, -pi/2, 0, pi/2, pi])
ax.xaxis.set_ticklabels([r'$-\pi$', r'$-\frac{\pi}{2}$', '0', r'$\frac{\pi}{2}$', r'$\pi$', ])

Everyone loves properly labeled plots.

[jtratner]

Can you post another image? :)

[jtratner]

Let's keep this to the colors for now and can deal with labeling
separately.

[amanahuja]

Just updated my last comment with an image. :) I think having the grid lines in multiples of pi makes a world of a difference.

[jtratner]

The colors change is minimal enough that I'm willing to merge it (and it seems clearly correct). We're in the middle of putting out a release so, in terms of changing labels or ticks, I'd prefer to hold off on that for a week or two. You can put up a separate PR that we can consider though.

[jtratner]

It looks like you have Travis set up but that it didn't build this. If you think Travis is set up can you run git commit --amend -C HEAD && git push --force to get Travis to notice the build?

[tacaswell]

I hadn't enabled it to build. Travis is running now.

[tacaswell]

And it looks like it built my master branch (which is just master).

[jtratner]

Need to check out the fix_andrews... branch

[tacaswell]

Sorry, I am really confused by your travis set up (I am used to mpl where PRs are auto-magically built by the matplotlib travis account).

[jtratner]

you did it :)

If there's a way to do that in Travis, we probably should - that would be much easier for contributors.

[jtratner]

Can you add a quick note to release.rst? After that I'll merge.

[tacaswell]

Done.

Squashed the commits.

[jtratner]

If you could, could you change the commit name to something like: "BUG: Fix color selection in andrews curve" and then you can put "fixed bug introduced in b5265a5
which change from picking a single random color for the class to
picking a random color for each line."

In the commit message? Makes it easier to follow what's going on.

[tacaswell]

like that?

[jtratner]

exactly. once it passes Travis I'll merge - thanks for the final cleanups!

[jtratner]

thanks!