Add plot() function

[mnlevy1981]

@andersy005 or @mgrover1 -- I'm not totally sure where to start with this one, and maybe I'm overthinking it... but I'm getting caught up in how overloaded xr.DataArray.plot() is, so I don't really know what the desired output of this function is. Suppose we have

import xarray as xr
import xcollection as xc
ds = xr.tutorial.open_dataset('rasm')
dsa = xr.tutorial.open_dataset('air_temperature')
collection = xc.Collection({'rasm': ds, 'air_temp': dsa})

How would you call collection.plot() and what would you expect the plot to look like? I was initially picturing something like

def plot(var, *args, **kwargs):

but collection['rasm']['Tair'].plot(); is a histogram and you need collection['rasm']['Tair'].isel(time=___).plot(); to get a contour plot. So maybe

def plot(var, isel_dict={}, *args, **kwargs):

and we could call

collection.plot('Tair', {'time': ___})

but then what happens if multiple datasets in the collection contain var? Should we allow collections to containmatplotlib.figure.Figure objects? Then collection.plot() could return a collection of figures?

That seems like useful behavior for contour plots, but what about line plots? If I want to look at a time series or vertical profile at a given location (or maybe of a global mean), then should there be an option to return a single figure containing lines from all acceptable datasets? E.g. if I have a collection of 10 datasets, and five of them contain the variable Tair, then maybe I want to plot all five together with a legend containing the key for each dataset?

[jukent]

I think I want to look more into how xarray.DataSet.plot() works. How does that plot method determine if we have a line or contour plot? And since when you have multiple time steps, for a contour plot it only plots the first one -- is it more similar to only plot the first DataSet in our Collection (with a kwarg for changing which DataSet it is? I am interested to hear @andersy005 and @mgrover1's thoughts.

[mnlevy1981]

Another design question: do we want to accept Collection.plot([var1, var2])? If so, does that mean returning a dictionary of dictionaries? Right now, Collection.plot(var) returns a dictionary of plots: return_dict[key] = <matplotlib.figure.Figure> but maybe it should be return_dict[key][var] = <matplotlib.figure.Figure> to support multiple variables at a time? I like the similarity between the existing .plot() results and the Collection structure but don't see how to keep that and allow plotting multiple variables at a time.