Currently the MRT2step adjustment and decorrelation is handled automatically, which leads to a variable amount of non-sampling steps before actual sampling occurs. This is not critically problematic in single thread use, but very undesirable in parallel use.
To solve this:
allow to specify a fixed number iterations/steps for both routines
retain the old integrate interface as wrapper for an easier (true/false) configuration and compatibility to existing code
add another unittest to test that decorrelation is still successful when using fixed values
Ithanil
commented
6 years ago
Currently the MRT2step adjustment and decorrelation is handled automatically, which leads to a variable amount of non-sampling steps before actual sampling occurs. This is not critically problematic in single thread use, but very undesirable in parallel use.
To solve this: