slope_aes

[Mattriks]

Fixes #1200.
This fix will work for transformations of the y-variable. As yet untested for transformations of both variables.

• I've updated the documentation to reflect these changes
• I've added an entry to NEWS.md
• I've run the regression tests
• I've built the docs and confirmed these changes don't cause new errors

Here's my log-linear example from #1200. In both plots, the intercept=21 and slope=-0.2.

Random.seed!(123)
 x = 15*rand(20)
 D = DataFrame(x=x, y= exp.(-0.2*x.+3.03) .+ 0.2*randn(length(x)))

abline = Geom.abline(color="red", style=:dash)

pa = plot(D, x=:x, y=:y,  Geom.point, Scale.y_continuous,
      intercept=[21], slope=[-0.2],  abline )
pb = plot(D, x=:x, y=:y,  Geom.point, Scale.y_log,    
     intercept=[21], slope=[-0.2], abline )

draw(PNG(6inch,3inch), hstack(pa, pb))

[Mattriks]

I see the Travis build is failing: same error as JuliaStats/RDatasets.jl#61.

[tlnagy]

I think this is caused by recent changes in CSV.jl. Does it pass locally if you pin the version of CSV to v0.3.1?

[Mattriks]

Here are examples of transformations of both y and x:

Random.seed!(123)
 x = [20*rand(20); exp(-3)]
Da = DataFrame(x=x, y= (-0.5*sqrt.(x).+5).^2 .+ 0.1*randn(length(x)) )  
Db = DataFrame(x=x, y= exp.(-0.5*asinh.(x).+5) .+ 0.1*randn(length(x)) ) 
Dc = DataFrame(x=x, y= (-0.5*log.(x).+5).^2 .+ 0.1*randn(length(x)) ) 

abline = Geom.abline(color="red", style=:dash)

pa = plot(Da, x=:x, y=:y,  Geom.point,  Scale.x_sqrt, Scale.y_sqrt,
      intercept=[25], slope=[-0.5],  abline, Guide.title("T(x)=√x") )
pb = plot(Db, x=:x, y=:y,  Geom.point,  Scale.x_asinh, Scale.y_log,
     intercept=[148], slope=[-0.5], abline, Guide.title("T(x)=asinh(x)") )
pc = plot(Dc, x=:x, y=:y,  Geom.point,  Scale.x_log, Scale.y_sqrt,
    intercept=[25], slope=[-0.5], abline, Guide.title("T(x)=ln(x)") )
hstack(pa, pb, pc)

It all works (e.g. plots a and b), with the caveat that plot c doesn't work in the sense that the intercept=25 doesn't match the intercept at x=0 (because log_(x) is asymptotic). This is a minor caveat: the slope in plot c is correct, and 25 is the intercept at x=1 rather than x=0. I've updated the docs to tell users about the minor caveat.

[Mattriks]

Yes the tests pass locally with CSV 0.3.1.

[tlnagy]

Yes the tests pass locally with CSV 0.3.1.

I think we should go ahead and merge because those changes aren't related to this PR

[Mattriks]

Ok!