- our wide uses of the tabulate procedure
- the need for shipping data across frameworks
- proc tabulate gives us a "realized" table with all nestings computed
- but how do we communicate and store the logic of pivots that leads to each table cell?
- we use a variety of languages as glue across our projects (Python, some R, node.js, etc)
- the unique nature of
proc tabulateis that its concise definition of nesting logic is not available in any other framework
- the nested logic of table statement (concatenation versus
*operator) can be parsed readily by any tokenizing system - in our case, we use
peg.jsto render a simple syntax tree that duplicates the nesting in a returned object - understanding how to bulid and traverse this tree makes the basic principles of
proc tabulatereadily apparent:- indices of
classvariables are built into a nested map of your data rows, by axis - concatentation allows more than one route at the same level
- a cell in the table represents the union of the sets produced by the left and top axes intersecting at that point
- set operations are executed against a single value selected from that unioned set
- indices of
- by connecting the procedure's output to the object describing the tabulation statement used, we can query specific subsets by named paths and return the value calculated by SAS
- in the interest of showing the value of work across paradigms, I will also demonstrate our method of recalculating basic statistics in nested groups using the output tree and Facebook's GraphQL syntax