Seamless qualitative variable support
ralsei opened this issue · comments
Currently, renderers that should have support for qualitative variables do not. While some renderers only make sense for quantitative variables (lines
for a time series, histogram
, ...), the notable exceptions are points
and col
.
Namely, points
should be able to display a dot-plot, and col
should achieve similar results to bar
(except the data should be taken literally).
Furthermore, boxplot
should not assume a qualitative variable. In short, we need qualitative-quantitative inference. By all means, this "inference" could just be string?
Note that since plot
generally expects that both axes are quantitative (and anything against it requires some isomorphism to a subset of the reals), this is antithetical to the design of plot
.
Partially resolved by #6. There are a few things I want to work out, namely:
- More extensive testing, with transforms and facets
- Disjoint axes while faceting (so the qualitative axis is duplicated, and filtered to be in each facet)
- Quantitative box-plots
- Errors when given quantitative/qualitative variables where one is not expected
Altogether, with define-renderer
etc, metadata, and now dealing with qualitative and quantitative variables, I'm beginning to get something kind of close to ggproto
and that functionality, at least internally. It may be useful to look at what ggproto does here, and its semantics.