* Added "light jumps" random walk algorithm. The new algorithm chooses
differently which next candidate point (for jumping to): instead of
exploring all the neighboring points uniformly on the hypersphere, we only
go to neighboring points in directions corresponding to the basis vectors.
This is much faster for large system sizes, and explores the same
distribution.
* Added support for *random walk controllers*, controlling for instance the
random walk step size, or controlling its duration to make sure that all
error bars calculated from a binning analysis have converged. These
controllers are implemented both in `tomorun` and in Python's
`tomographer.tomorun.tomorun()`
* `tomorun` program: New internal structure for defining figures of merit, it
is now much more easy to add a new figure of merit
* `tomorun` program: Can now seed the random number generator via a system
random device (requries custom compilation)
* Updated documentation
* Various fixes to the build system
* Added more minimal examples (`test/minimal_tomorun_controlled.cxx` &
`test/minimal_single_random_walk.cxx`)
* Added `ChangeLog.md`
C++ API Changes
* Added `Tomographer::DenseDM::TSpace::LLHWalkerLight` class for "light
jumps", which is much faster for large system sizes
* Added the *MHRWControllers* concept, and implementations
`Tomographer::MHRWStepSizeController`,
`Tomographer::MHRWValueErrorBinsConvergedController`, as well as
`Tomographer::MHRWAcceptRatioWalkerParamsController`
* `Tomographer::UniformBinsHistogram` → `Tomographer::Histogram` (etc.),
deprecated aliases available
* Added `Tomographer::AggregatedHistogram{Simple|WithErrorBars}` classes
* Some API clean-ups
**Major Backwards-Incompatible API Changes to Some Components:**
* Changed `Tomographer::MHRWParams` template parameters, now allowing for
arbitrary *MHWalkerParams* specifying whatever parameters the MHWalker needs
in order to carry out its random walk
* Status reporting API changed. Introduced
`Tomographer::Tools::StatusProvider` and `Tomographer::Tools::StatusQuery`.
Now the `MHRandomWalk` is itself aware of status reporting. Added
`Tomographer::PeriodicStatusReportMHRWStatsCollector` and
`Tomographer::PredStatusReportMHRWStatsCollector`.
* `Tomographer::MHRWTasks::ValueHistogramTasks` was removed and replaced by
`Tomographer::MHRWTasks::ValueHistogramTools` with a new revised API with
much improved flexibility. (For instance, it is now possible to have several
stats collectors and to customize what results on is interested in.)
* `Tomographer::MHRWTasks` was reworked for added flexibiliy (e.g., the
*CData* now needs to specify `setupRandomWalkAndRun()` etc. instead of
`createMHWalker()` and friends, so that it can add controllers and keep
references to each other etc.)
Python API Changes
* New histogram classes with NumPy buffers, replacing the simple C++ wrappers.
Renamed `tomographer.UniformBinsHistogram` → `tomographer.Histogram`,
removed the different classes for different underlying types (now all
handled by `tomographer.Histogram`). Deprecated aliases available.
* Deprectated the `AveragedXXXXXXHistogram` classes
* `tomographer.MHRWParams` now stores any arbitrary *MHWalkerParams* using an
arbitrary Python object (normally a dictionary). It is backwards-compatible
in several ways (MHWalkerParams as additional keyword arguments, or a step
size given as first argument to MHRWParams())
* (python) added compiler info in `tomographer.version.compile_info`
* Doc: Example for writing your own C++ code with a Python module interface
* `PyLogger` made explicitly not GIL-aware, use `GilProtectedPyLogger` if GIL
is released
* Added test cases (e.g. unpickling data pickled using earlier versions of
Tomographer)