- More tests.
- Changed SparseStateRewardParser to a static class
- Added comments here and there
- Some reformatting.
- Fixed some warnings.
- Eliminated some unnecessary includes.
- ...
Former-commit-id: efe1c96fee
- Noticed to my astonishment that seemingly arbitrary use of whitespaces (as long as each transition is in its own line) is no problem for the parser.
- As predicted, handling of action labels of arbitrary length especially such starting with an '!' is not supported.
Next up: Handle arbitrary labels.
Former-commit-id: 339578e72a
- gcc 4.8 (and assorted libraries) does not provide an erase(const_iterator) method for std::list but only an erase(iterator).
This is in compliance with the c++11 draft N3337, which specifies the change from iterator to const_iterator only for "set, multiset, map [and] multimap" but not list.
Therefore the constant list iterators were replaced by non constant iterators in MaximalEndComponentDecomposition and Vector set.
The locations are marked with a FIXME, such that the const_iterator can be replaced back when gcc provides it.
- Fixed (completed) the stub implementation for the GurobiLpSolver in case that Gurobi is not present.
|-> Would not compile before due to missing functions and incorrect signatures.
- Switched to c++11 for gcc. Since gcc 4.8 provides full compliance to the c++11 standard.
|-> Initially hoped that it would fix the const_iterator problem, but it did not.
- Fixed the cmake warning concerning a missing whitespace between tokens in the last line of CMakeLists.txt.
Former-commit-id: f90768375e
- Fixed one problem marked FIXME in the transpose function. The need for a "sentinel" element was created by an off by one in the prior loop.
- Changed all occurences of SparseMatrix<bool> to SparseMatrix<T>. Now the only two types for which SparseMatrix is instantiated are double and int.
- Compiles again.
|-> Compile time seems to be roughly the same for clean; make all. For incremental builds I haven't tested yet.
Former-commit-id: 6d829e0903
But got bitten by std::vector<bool> as it is specialized and uses bitsets (i.e. integers) internally.
Less memory but at the cost of 'oh, sorry std::vector<bool> does not return a bool&'.
That again seems to be a problem for the SparseMatrix<bool> instatiation since for instance getValue returns a T&.
On the one hand I don't quite know why this was never an issue before and on the other hand it prevents successful compilation.
So there are different ways to settle this:
- Specialize SparseMatix for bool -> possibly lots of code, but might be the best solution
- Write a wrapper for std::vector that uses chars instead of booleans
- Dont't use SparseMatrix<bool>
Next up: Figure out the best solution for this and implement it.
Former-commit-id: 83b9cfd06e
masawei
dehnert
David_Korzeniewski