Support iris.plot.hist and iris.quickplot.hist

[bjlittle]

There have been user requests for the iris convenience plotting capabilities to be extended such that a cube can be passed to iris.plot.hist and/or iris.quickplot.hist to generate histogram plots by using matplotlib.pyplot.hist

@BenMGeo do you have some example use cases to support his issue?

[pelson]

Hi from the other side! I found myself needing something like this recently... here is my use case:

I have all of the observed temperatures for a given site, I take only those of a given month and hour, and want to see the distribution of temperatures for that hour in that month. I then plan to fit that distribution to a statistical model and use it to calculate the climatological probability of a given temperature at a given time in a given month (pretty useful!).

[pelson]

For my own record, I hacked something together:

n_bins = 20
counts, bins = np.histogram(temp.data, np.linspace(temp.data.min(), temp.data.max(), n_bins + 1))
bounds = np.vstack([bins[:-1], bins[1:]]).T

frequency_cube = iris.cube.Cube(
    data=counts, long_name='temperature_frequency', dim_coords_and_dims=(
        (iris.coords.DimCoord(points=bounds.mean(axis=1), bounds=bounds, long_name='temperature'), 0),
    ),
)

And then I did a bar plot of the cube with:

# Plot the data as a histogram.
bar_widths = np.diff(frequency_cube.coord('temperature').bounds, axis=1)[:, 0]
plt.bar(frequency_cube.coord('temperature').points, frequency_cube.data, width=bar_widths)

Truth be told, this is probably a good use case for something like pandas. The only way I could convince myself that it would be a good thing for Iris to have is if we needed multi-dimensional equivalents. For example, imagine having all sites, and all months, and all hours as 3 distinct dimensions, and wanted to do the same analysis and resulting in a multidimensional cube.

[trexfeathers]

Truth be told, this is probably a good use case for something like pandas. The only way I could convince myself that it would be a good thing for Iris to have is if we needed multi-dimensional equivalents. For example, imagine having all sites, and all months, and all hours as 3 distinct dimensions, and wanted to do the same analysis and resulting in a multidimensional cube.

@pelson we've recently enhanced iris.pandas.as_data_frame(), so that it can handle the n-dimensional case. Would that suit future use cases for you?