Hierarchical Formats
No gods, no masters.
–Louis Auguste Blanqui
When we utilize machine learning models, and indeed when we perform general statistical analyses, we almost always assume our data is tabular. Observations and features; rows and columns. And yet, there are a number of very popular ways of storing data that resemble trees rather than plowed fields. Data objects belong to other data objects which belong to yet other data objects, with no specific limit on the depth or names of branches. Both for economy of understanding and, in the case of database systems, for efficiency of access, hierarchical data formats very often make more sense for a broad class of data.
There are many domains that simply map more naturally to hierarchies than to tables. Yes, the relational algebra—the conceptual structure that underpins SQL and relational databases—is in some way able to represent every possible structure. But it feels awkward for...
