Refactoring

[frapac]

• add varying bounds for states and controls
• fix HereAndNow SDDP
• add iterate! function to ease dual SDDP code
• drop julia 0.5 support

[odow]

When do we get to see the dual code? :) I'm very interested

[frapac]

I was planning to release it after the publication of the article on arxiv, but I have been a bit busy since :p

If you are interested, I can tidy up the code and release it by the end of the week!

[odow]

No rush, you have your thesis to finish!

[frapac]

@odow I released dual SDDP in this repo:
https://github.com/frapac/DualSDDP.jl

To use dual SDDP, you should be on the branch fp/refact in StochDynamicProgramming.

I give some example in a jupyter notebook. The code is not that clean, but I hope it conveys well the ideas behind dual SDDP :)

I think that writing the dual SP should be straightforward with MOI, as we have access to the Linear Bellman operator coded directly as a LP and we have access to the corresponding matrix. I plan to implement this after the JuMP dev workshop, in a new version of SDDP in JuliaStochOpt.

Do not hesitate if you have any remark concerning the code/the algorithm!

[odow]

I'm very pleased to see this. Great work. I'm submitting my thesis Tuesday so will look properly after then.

[frapac]

Good luck for the final rush!