Add MultiPanelFITSFigure class?

[cdeil]

I would like to add some code to aplpy to make it easy to plot very wide survey images in multiple panels, like in this example.

I do have some (very messy) code to do it where the user has to sub-class to make a plot:
https://gist.github.com/cdeil/9090342

Would that be a welcome addition or is this too specialized?
If yes, what would be a nice API for this?

The current implementation seems overly complicated, plus I think it's very slow and memory intensive because for each panel there is a copy of the full survey image data ... any advice how to do this better?
#88 might help.

[keflavich]

I would welcome it. This has been a common need in my daily work; last time I did this I think I did it in IDL!

Maybe it would be best to have a wrapper that splits up the FITS file into sub-files to minimize memory usage? This would remove some of the flexibility aplpy provides, but would improve performance.

Alternatively, maybe there's no need to do anything beyond #88, but instead include this in https://github.com/aplpy/aplpy-examples/ as a recipe?

I haven't looked at #88 in a long time; @cdeil, any chance you want to help add some tests?

[astrofrog]

I think this would be useful to have at the very least in http://aplpy.readthedocs.org/en/v0.9.11/howtos.html - whether it should make it into the core package is not as clear, but if you could already write a HOWTO with the simplest possible code, then we can see how we can generalize it to include it in the actual package.

[cdeil]

I still haven't figured out how to structure code for plotting multiple sky images on one matplotlib Figure well.
The issue @adonath and I are having is that we want to write one function per sub-plot (plot_ax_1 and plot_ax_2 in the example below) and then combine these sub-plots in various ways (e.g. make_plot_1 in the example below).

import matplotlib.pyplot as plt

def plot_ax_1(fig, rect):
    ax = fig.add_axes(rect)
    ax.plot([1,2,3], [2,3,3], color='blue')

def plot_ax_2(fig, rect):
    ax = fig.add_axes(rect)
    ax.plot([1,2,3], [2,3,3], color='red')

def make_plot_1():
    fig = plt.figure(figsize=(10, 10))
    ax1 = plot_ax_1(fig, rect=[0.1, 0.2, 0.3, 0.4])
    ax2 = plot_ax_2(fig, rect=[0.5, 0.6, 0.1, 0.2])
    fig.savefig('/tmp/plot_1.png')

    return fig

fig = make_plot_1()

Is this the best way to structure such code, take a matplotlib Figure and subplot rect as input, or

• would it be better to pass a dummy Axes and then replace this somehow with a WCSAxes?
• not pass anything in, but let plot_ax_1 create an Axes from scratch (possible without a Figure?) and then the Figure is created and the subplots placed later?

[astrofrog]

@cdeil - just to check, is this something you can do with WCSAxes? For instance:

import matplotlib.pyplot as plt

def plot_ax_1(fig, rect):
    ax = fig.add_axes(rect, projection=WCS(...))
    ax.plot([1,2,3], [2,3,3], color='blue')

def plot_ax_2(fig, rect):
    ax = fig.add_axes(rect, projection=WCS(...))
    ax.plot([1,2,3], [2,3,3], color='red')

def make_plot_1():
    fig = plt.figure(figsize=(10, 10))
    ax1 = plot_ax_1(fig, rect=[0.1, 0.2, 0.3, 0.4])
    ax2 = plot_ax_2(fig, rect=[0.5, 0.6, 0.1, 0.2])
    fig.savefig('/tmp/plot_1.png')

    return fig

fig = make_plot_1()

I'm still trying to figure out where to draw the line between WCSAxes and APLpy... and I need to understand the various use cases better. You should be able to do anything with WCSAxes but it will be a little more manual in some cases. For what you want to do here it seems you want to be close to the Matplotlib API so APLpy doesn't seem appropriate.

[cdeil]

@astrofrog – Here's an example of what I want to do:
http://nbviewer.ipython.org/gist/cdeil/3d6612a98aaf986cf60a

Show the same image twice for different zoomed in regions.
At the moment I'm trying with APLPy, because I wasn't sure how to to recenter with wcsaxes.

I would like to have the two panels have the same height ... do you know how I can prevent FITSFigure.add_colorbar from reducing the height of the image?

[astrofrog]

@cdeil - for WCSAxes you can just use set_xlim and set_ylim giving it pixel coordinates, and there is no way to zoom in in world coordinates (yet).

[keflavich]

@cdeil @astrofrog for the colorbar issue, see #119 (which is about colorbars & resizing but doesn't say so in the title)