piccolbo / altair_recipes

A selection of statistical graphics for vega in python, based on altair.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

develop charts for functions

piccolbo opened this issue · comments

The case of a single y value for each x (a finite function) has not been treated in any specific way. A scatter plot or stripplot will be used whether the data comes from a function or a relation. We can fix that.

  1. Develop a test to decide whether scatter or line chart is more appropriate and implement it in autoplot
  2. Implement charts that are used for functions such that line plot, area plot, stacked area plot and what not. The goal as always is not to cover everything but only the classics, and, at a minimum, what is needed by autoplot
  3. Add logic to autoplot to choose an appropriate graph to represent a function. We expect this to be of any interest only for discrete-> continuous and continuous to continuous case, replacing the strippplot and scatterplot with a barchart and line plot resp.
  1. is done as part of warn_not_distinct function, unless we want something more sophisticated
  2. is done in #11 and #12
    1 and 2 already merged.

What I mean by a sophisticated test is that the current test just looks for exact repetition in the x cordinate. If we have to plot one continuous random sample against another independent continuous random sample, the test would not detect any repetition but a sensible analyst would choose scatter over line. So we need to think further before this is incorporated in autoplot. Given also the complexity of the logic in autoplot, we may want to look into a reorg of that function to scale to more plot types before we add more. We'll release what we have for now.