matplotlib scatter shifts color codes when NaN is present

[Morisset]

I want to plot some data with a given color code. For example:

x = np.array([0, 1, 2, 3])
b = np.array([1, 0, 2, 3])
colors = np.array(['g', 'r', 'b', 'y'])
plt.scatter(x, b**2, color=colors)

Great, all the points appear, each one with its color. Now, some data is missing:

plt.figure()
plt.scatter(x, np.log10(b), color=colors)

Here comes the problem: the x=1 data is missing (log(0)=NaN), but the colors are skipping this point and the x=2 point is red, not blue. The solution can be:

y = np.log10(b)
mask = np.isfinite(y)
plt.scatter(x[mask], y[mask], color=colors[mask])

but I feel it so uncomfortable to do this... Any other way? Is this a kind of bug?
Thanks,
Christophe

[efiring]

Yes, it's a bug; but there is a function call in scatter that is supposed to be handling exactly this situation, and I don't see immediately why it is failing to do so in this instance. It looks like calling the function, cbook.delete_masked_points(), directly with the arguments works as intended, so it appears that something is getting fouled up prior to the function call.

[efiring]

Aha! @tacaswell was right--the problem was the API. The correct kwarg is "c", not "colors". I don't know what to do about that; maybe pop "colors" off the kwarg list and use it as a synonym for "c"?

[tacaswell]

Right by dumb luck!

And that sounds reasonable to me use color as an alias for c.

In fact right now you can pass in both c and color and in some cases it looks like color will over-ride c.

It also looks like facecolor, facecolors, edgecolor and edgecolors will all also have similar problems.

[edit to finish sentence]

[WeatherGod]

All the more reason for some sort of kwarg normalization function discussed
yesterday...

On Tue, Sep 9, 2014 at 11:01 PM, Thomas A Caswell notifications@github.com
wrote:

Right by dumb luck!

And that sounds reasonable to me In fact right now you can pass in both c
and color and in some cases it looks like color will over-ride c.

It also looks like facecolor, facecolors, edgecolor and edgecolors will
all also have similar problems.

—
Reply to this email directly or view it on GitHub
#3489 (comment)
.

[story645]

does #4079 also fix this one?

[efiring]

@story645 I suspect it does, possibly in conjunction with later changes. Would you test v2.x with the example given above, please? If you find that the bug no longer appears, I will close this issue.

[story645]

Sorry had totally spaced out on this. Tested and still broken, but now there's #7570

[tacaswell]

@story645 Could you review #7570 ?