List view
The stochastic dynamic programming part of the package
No due date•3/3 issues closed- No due date•1/3 issues closed
One of the use of the package is to allow easy benchmarking of different approaches and choice in SDDP. So part of the project should be devoted to easily benchmark different choices of SDDP. - For a given SPmodel, use a fixed list of noise scenarios for the forward phases allowing -- comparison of solvers (parameters) efficiency -- benchmarking of cut-selection methods in term of time / memory -- benchmarking of approximate-solution efficiency in term of time -- comparison of the quality of the solution obtained with different choices of SDDP parameters - For a given SPmodel and SDDPparameters keep in memory lower bounds (and maybe upper-bounds) - For a given problem compare different risk-aversion choices
No due date•0/3 issues closedEnable a feasability cut procedure
No due date- Cut pruning - epsilon solution
No due date•1/1 issues closedImprove the algorithm efficiency. Different approaches : - julia profiling - (asynchronous) parallelization - Linear modelling (especially considering to have or not the artificial constraint x=x_t^k used to get the multiplier) - Initialization
No due date•0/1 issues closedAllow for nested risk measures in the objective function. Risk measures can be defined by : - extreme points of dual - convex combination of expectation and AVAR
No due dateAdapt the code to allow convex-(piecewise)quadratic costs and constraints with linear dynamics.
No due dateThis part of the project address the problem of communication between an user and the package. Indeed giving by hand every dynamic constraint is sub-efficient. The objectives : - Easily create stock dynamics - Easily create stock interactions - Easily integrate random modification of the dynamics (like outage) - Allow for a user-defined AR-process, the package building the linear dynamics and the discretization of the residual.
No due date•5/7 issues closedThe objective is to have a tested core project to present to the JuliaOpt community in order to have feedback on the code. Features required : - the user define : -- linear or polyhedral cost functions -- linear dynamics -- independent noise law -- initial state and control bounds - the package allow then to -- construct an extensive formulation and call a solver on it -- compute through SDDP a lower bound, an estimation of the cost and a probabilistic gap - We have (at least) two examples : -- a simple, easy to understand, dam problem -- a randomly generated problem to test efficiency - We have some documentation for the project core
Overdue by 9 year(s)•Due by March 14, 2016•9/9 issues closed