Tim Quatmann
|
6891825803
|
IterativeMinMaxLinearEquationSolver: Fixed not incrementing an iterator when computing the maximum absolute difference between two values
|
5 years ago |
|
Tim Quatmann
|
bc623d1203
|
MinMaxLinearEquationSolver: Added a flag 'hasNoEndComponent' that is true if the system is known to have no end components. This decides if policy iteration does require a valid initial scheduler.
Renamed the 'hasNoEndComponents' solver requirement to 'hasUniqueSolution' as this is the actual thing we require for, e.g. sound value iteration.
|
6 years ago |
|
Tim Quatmann
|
5467043807
|
DeterministicSchedsLpChecker: Only consider end components with value zero for all objectives.
|
6 years ago |
|
TimQu
|
6e3639c8f1
|
Added new minmax method: Vi-to-Pi, which first performs value iteration with doubles, to find a good initial policy for (potentially exact) policy iteration.
|
6 years ago |
|
TimQu
|
94fb16e654
|
svi now considers bounds by default
|
7 years ago |
|
TimQu
|
12f8685080
|
Custom Termination Conditions for sound value iteration
|
7 years ago |
|
TimQu
|
8b00f8441e
|
Improved caching for svi
|
7 years ago |
|
TimQu
|
a24de86ce1
|
Avoided duplicated code for sound value iteration
|
7 years ago |
dehnert
|
7150354b9d
|
fixing issue related to vector swapping in (explicit) value iteration and power method
|
7 years ago |
|
dehnert
|
acf297a811
|
fixing precision issue in sanity check and silencing min-max solver a bit
|
7 years ago |
|
TimQu
|
24382630dc
|
removed output of performed iterations to cout
|
7 years ago |
|
TimQu
|
c1ecc22303
|
new multiplyRow method for sound vi
|
7 years ago |
|
TimQu
|
e491dc3813
|
fixed usage of multiplyrow
|
7 years ago |
|
TimQu
|
48945d1199
|
improved multiplyRow method
|
7 years ago |
|
TimQu
|
b7bac59ae0
|
Using multiplier in IterativeMinMaxSolvers
|
7 years ago |
|
TimQu
|
09bcaa8db8
|
fixed correct requirements in minmaxsolver
|
7 years ago |
|
TimQu
|
ea25c8fd2e
|
renamed 'sound value iteration' to 'interval iteration' and 'quick sound value iteration' to 'sound value iteration'
|
7 years ago |
|
TimQu
|
f168df139d
|
made qvi code more readable
|
7 years ago |
|
TimQu
|
674a30c154
|
removed some 'experimental' code and introduced the #iteration output again
|
7 years ago |
|
TimQu
|
ca1bcebc71
|
Added an option that enforces updating both bounds in interval iteration
|
7 years ago |
|
TimQu
|
13184aefcf
|
Printing the number of performed iterations (reverted from commit b5399d0596)
|
7 years ago |
|
TimQu
|
cebf29ef23
|
trying something else
|
7 years ago |
|
TimQu
|
26362ed36f
|
trying an alternative implementation of qvi
|
7 years ago |
|
TimQu
|
b5399d0596
|
Printing the number of performed iterations
|
7 years ago |
|
TimQu
|
a2bd1e0026
|
renamed argument from getRequirements so that it is easier to understand
|
7 years ago |
|
TimQu
|
3b394a965e
|
some qvi optimizations
|
7 years ago |
|
TimQu
|
8c3991fb2f
|
respecting lower/upper bounds from preprocessing in quick sound power method
|
7 years ago |
|
TimQu
|
78cfb10c7e
|
fixed qvi with negative rewards
|
7 years ago |
|
TimQu
|
fc422af557
|
making things compile again in debug mode
|
7 years ago |
|
TimQu
|
fa7f74f0f1
|
quicker iterations when the decision value blocks the bound
|
7 years ago |
|
TimQu
|
0215258709
|
made qvi implementation a little bit more readable
|
7 years ago |
|
TimQu
|
c81c7b0be5
|
Fixed issue with restarting
|
7 years ago |
|
TimQu
|
80219e4a2d
|
quick value iteration restart
|
7 years ago |
|
TimQu
|
7aac41d8f2
|
optimized qvi implementation
|
7 years ago |
|
TimQu
|
116bd58b22
|
log improvements + minor bugfixes for qvi
|
7 years ago |
|
TimQu
|
9c96bd0a1c
|
First implementation of quick value iteration for MinMax Equation systems
|
7 years ago |
|
TimQu
|
1cff0fcbbb
|
improved interface of solver environment
|
7 years ago |
|
TimQu
|
68ec4ca0ce
|
Various fixes for the case STORM_USE_CLN_EA=ON
|
7 years ago |
|
TimQu
|
776ce4c8bb
|
Checking requirements of a linear equation solver now depends on whether we want to do multiplication or equation solving. This was necessary to get the correct requirements of a MinMaxSolver that only uses the underlying linear equation solver for multiplication.
|
8 years ago |
|
TimQu
|
e09cb86001
|
making sure that the default linear equation solver is not switched to native if we check e.g. an MDP with sound value iteration
|
8 years ago |
|
TimQu
|
e46f4e154b
|
1. Ensured that when doing policy iteration the underlying solver is at least as precise as the minmax solver.
2. Fix for SolverGuarantees: They are only established if bounds were actually given.
|
8 years ago |
|
TimQu
|
bb63ac6089
|
Linear equation solver + game solvers now respect the environment as well
|
8 years ago |
|
TimQu
|
25c006ec13
|
fixed method selection of iterative min max solver
|
8 years ago |
|
TimQu
|
42cea9c688
|
better subenvironments
|
8 years ago |
|
TimQu
|
6d23c79737
|
Making libstorm compile again
|
8 years ago |
|
TimQu
|
fd8c99b989
|
Introducing Environment in MinMaxSolvers and ModelCheckers
|
8 years ago |
|
TimQu
|
d85a845b7d
|
Fixed cases where a solver guarantee was not established although it was needed by the termination condition
|
8 years ago |
|
TimQu
|
3571f0ddca
|
Respected that the solution is unique when doing value iteration
|
8 years ago |
|
TimQu
|
33585c811f
|
MinMax Solver requirements now respect whether the solution is known to be unique or not.
|
8 years ago |
|
dehnert
|
d90c507431
|
fixed bug in sparse bisimulation quotient extraction related to rewards
|
8 years ago |
