wssbck / crossfilter-helpers Goto Github PK

ReduceRemove (or rem) for aggregations by min and max return incorrect values.

Due to impressions in floating point arithmetic, even simple reduction by sum can behave incorrectly when a dimension is filtered to one entry and calculations are done using standard JS Number floats. For example, given the dataset:

    data  = [
        { id :  1, g : "A", n :   10 },
        { id :  2, g : "A", n :    4 },
        { id :  3, g : "A", n :    2 },
        { id :  4, g : "A", n :    0 },
        { id :  5, g : "A", n :    2 },
        { id :  6, g : "B", n :    0 },
        { id :  7, g : "B", n : 0.15 },
        { id :  8, g : "B", n : 0.15 },
        { id :  9, g : "B", n :  0.1 },
        { id : 10, g : "B", n :  0.2 }
    ]

If we grouped by dimension 'g' and reduced by sum of 'n', then filtered dimension 'id' to 6, we would not receive 0, but a tiny value close to it.

Perhaps instead of supporting decimal.js only, there should be a way to supply one's own arbitrary precision library?

1. Decide on the treatment of null values (preferably meaning "lack of data" or "result impossible to compute")
2. Substitute nulls for initial values (breaking change!)

Parameter: comparison function to determine the first and last element of a group.