[Test] Setup test for `add_rank_size`

[mmcky]

Setup a test for function contributed via PR #518

[shrabian]

Hey I'd like to take this issue

[bktaha]

As the author @jstac mentions in the PR, the method is written as it is so as not to add new external dependencies ( matplotlib ). It takes distribution data and an matplotlib.Axes object as parameter, generates rank_data and size_data and in turn passes those to the Axes object. So there is no return variable, only the Axes object with a changed state. Example usage from docstring :

import matplotlib.pyplot as plt
y = np.exp(np.random.randn(1000))  # simulate data
fig, ax = plt.subplots()
rank_size_plot(y, ax)
plt.show()

In a test then we will need to use get methods of the Axes object to retrieve rank-size data, and verify against a standard output. As I understand this will require adding matplotlib as a dependency, which the author was trying to avoid in the first place.

--

An alternative could be to rewrite this method to take only the distribution data and return rank-size data, using a similar signature as the lorenz_curve() method in the same module. Setting of log scale on the axes, labels and other parameters will have to be done separately like so,

rank_data, size_data = rank_size(y)
ax.loglog(rank_data, size_data, 'o', markersize=3.0, alpha=0.5, label=label)
ax.set_xlabel("log rank")
ax.set_ylabel("log size")
plt.show()

Do the maintainers have thoughts on how to go about this?

Apologies for the long explanation, I tried to convey my approach as I see it. Please add if I am mising something.

[jstac]

@bktaha , I think this is an excellent proposal. It's a better solution and your explanation is very clear.

The only issue is that it will break some code in the lectures. You can find one example in the solution to exercise 2 here: https://python.quantecon.org/wealth_dynamics.html#Exercise-2

Hence it would be necessary to adjust the lectures and the code library at the same time.

The full list of changes that need to be made in the lectures can be found by cloning https://github.com/QuantEcon/lecture-python and searching for the string in /source/rst.

Then we can do a simultaneous merge, and a new release of the library (CC @mmcky).

Might you be willing to take the full task on @bktaha? Or might @shrabian be able to help? (Sorry we missed your message.)

[bktaha]

Based on this discussion, there are 3 such lectures.

• Heavy-Tailed Distributions
• Wealth Distribution Dynamics
• Kesten Processes and Firm Dynamics

I cross-checked this by searching for the string in the /source/rst directory.

--

I'd be happy to take the full task on.

Should I keep the name rank_size_plot or change it to rank_size or rank_size_curve ?

[jstac]

Thanks @bktaha, that's greatly appreciated.

I'll look forward to reviewing your PRs.

Let's go for rank_size.