Smart bounds for (s)pdmp

[mschauer]

No description provided.

[codecov-commenter]

Codecov Report

Merging #83 (d2358dd) into master (0b961bf) will decrease coverage by 8.69%.
The diff coverage is 29.90%.

@@            Coverage Diff             @@
##           master      #83      +/-   ##
==========================================
- Coverage   80.27%   71.57%   -8.70%     
==========================================
  Files          18       19       +1     
  Lines        1090     1256     +166     
==========================================
+ Hits          875      899      +24     
- Misses        215      357     +142     

Impacted Files	Coverage Δ
src/ZigZagBoomerang.jl	100.00% <ø> (ø)
src/local.jl	0.00% <0.00%> (ø)
src/trace.jl	67.39% <0.00%> (-23.79%)	⬇️
src/poissontime.jl	85.00% <50.00%> (+0.78%)	⬆️
src/types.jl	81.25% <60.00%> (-10.42%)	⬇️
src/not_fact_samplers.jl	77.64% <72.58%> (-18.19%)	⬇️
src/dynamics.jl	100.00% <100.00%> (ø)
src/sfact.jl	87.70% <100.00%> (-0.10%)	⬇️
src/ss_not_fact.jl	93.65% <100.00%> (ø)
src/staticarrays.jl	100.00% <100.00%> (ø)
... and 2 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update da5eebb...d2358dd. Read the comment docs.

[mschauer]

pdmp is now internally working with a multi argument version of the gradient which also returns the change of the gradient in direction of the trajectory to compute smart bounds.

ℓ(x, args...) = ...
function gradients(ℓ, d)
    function (y, t, x, θ, args...)
        x_ = x + Dual{:Bz0iEasFfw}(0.0, 1.0)*θ
        y_ = ForwardDiff.gradient(x->-ℓ(x, args...), x_)
        y .= value.(y_)
        y, dot(θ, y_).partials[]
    end
end
∇ϕ! = gradients(ℓ, d)

Also update trace for bouncy, add cummean.