Description
Binning continuous data into discrete bins is a non-trivial problem, especially when there are features of various scales that one is trying to capture. This astropy example is much more clear than any I could hope to write. The current implementation of pandas.cut does not allow for easy user access to these tools.
numpy provides a number of simple binning algorithms to capture these features in computationally efficient ways. As done in astropy, perhaps we can pass the bin edge calculations directly to numpy. To do so, we would need to call histogram_bin_edges (available in numpy 1.15.1, if not earler). This function isolates the calculation of bin edges.
To utilize np.histogram_bin_edges, a user would call
data = np.random.normal(100000)
edges = np.histogram_bin_edges(data, bins="auto") # Other strings are acceptable.
cut = pd.cut(data, bins=edges)
We can simplify this by replacing one or two lines in pandas.cut source
bins = np.linspace(mn, mx, bins + 1, endpoint=True)
becomes
bins = np.histogram_bin_edges(x, bins=bins)
so that the user can simply call
data = np.random.normal(100000)
cut = pd.cut(data, bins="auto") # Or any of the other strings accepted by numpy
and have access to all the techniques available there.
The histogram_bin_edges docs are availble here. The source is here.