Improving plots recipes

[sciml]

Improving plots recipes

[tbreloff]

So is there some reason why this pattern wouldn't work?

@gif for j=1:num_steps
    #  TODO update the atom progress
    timeseries = [ts[j] for ts in sol.timeSeries]
    plotfunc(sol.femMesh.node[:,1], sol.femMesh.node[:,2], timeseries)
end

And then you'd create either/both of "type recipes" and "user recipes" depending on what you need.

[tbreloff]

I've never actually tried switching backends mid-recipe. It might work???

[ChrisRackauckas]

I am still not quite sure how to make the @gif command be in a recipe. Do you just put it at the bottom, and so the recipe is returning an animation instead of a plot?

I was just following your example with the backend thing. The reuse=true is necessary, and if that's the case it's nice to have it in the animation call, but I am wondering if it could grab the current backend and add that, and then after the animation is made turn it back to the way it was?

[ChrisRackauckas]

FYI the tests fail after removing the checkout calls from the tests. Has the METADATA just not updated or something? It fails because the keyword plottrue that I defined for some recipes is not found, which is an issue only when called on an older version of Plots.

[tbreloff]

What are the types on these:

sol.femMesh.node[:,1],sol.femMesh.node[:,2],sol.timeSeries[i][j]

I am still not quite sure how to make the @gif command be in a recipe. Do you just put it at the bottom, and so the recipe is returning an animation instead of a plot?

The recipes would be called from within the plotfunc call. You wouldn't add any animation logic to a recipe.

You made me realize an important point (that I was already pondering but chose to ignore), that recipes aren't used when pushing/updating data. I think it would make sense to process "type recipes" inside push!/setindex! methods, but other recipes would be tricky to get right generically. If you're pushing data into a histogram recipe, for example, you'd want to update the bins accordingly, which involved changing the original data and then re-processing the recipe. This is certainly do-able, but might be a "0.8 feature".

[tbreloff]

Though looking at your animation function, it seems like you're just rebuilding a plot for each frame, so things would work fine.