Mdt

===================

Added
-----
- Adds infrastructure to be able to represent the gradient vector as spherical angles in a compute kernel.
- Added weight sum to one transformation to the FIM objective function.

Changed
-------
- Removed local reduction from the Tensor-DTI post-processing. This was unnecessary.
- Moved all the input data classes and functions to a dedicated module.
- Renamed get_parameter_codec to get_mle_codec.

Other
-----
- Followed changes in MOT allowing the work to be better splitted over workitems.
- Moved repository to private github account.

===================

Added
-----
- Adds optimization and runtime options to the get_optimization_inits function.
- Adds wild and residual bootstrapping functionality.
- Adds a post-processing flag to the model to disable computation of the log likelihood and the information criterion maps.
- Adds a separate compute_fim function to MDT. This can be used to compute the FIM matrix post-hoc.
- Adds the IVIM model.
- Adds the poisson distributed ActiveAx model, courtesy by Mark Drakesmith from CUBRIC.

Changed
-------
- Changed some of the parameter transformations.
- Changed signature of the input data copy_with_updates function.

Other
-----
- From the composite model, removed the voxels_to_analyze context. Instead, we now use the 'get_subset' method from the kernel data objects. This allows for cleaner code.
- Removed a few more mot_float_types for either float or double.
- Implements the meta-parameters 'observations', 'observation_ind' and 'nmr_observations' in the compartment model.
- Removed Numpy future warning concerning stacking using generators.

===================
Version 1.0 marking the end of the PhD of MDT's lead developer Robbert Harms.
This software package will still be supported for the foreseeable future by Robbert.
Future development will continue by CBCLab, Maastricht University under the supervision of Alard Roebroeck.
For ideas, please check the "Development Ideas" page.

As a personal note to all the users of MDT. Thank you all for using MDT and for your invaluable feedback over the past years. I could not have done it without you.

Best wishes, Robbert

====================
This version marks the complete removal of the Cascade models.

MDT still does cascaded initialization by default, but now using predefined initializations.
The automatic initialization can of course be disabled and or overwritten using manual provided data. See the manual for instructions.

There are two reasons for this change. First, the default cascade was used in 99% of the cases. Removing it and making it an implicit default simplifies the code.
Second, the code change provides an opportunity of future extensions towards multi-modal modeling.

Added
-----
- Added base class for EstimableModels.

Fixed
-----
- Fixed bug in handling the gradient deviations.

Removed
-------
- Removed the Cascade models.

Other
-----
- Slight restructuring of the modules.

====================

Changed
-------
- Reverted the parameter transformation of the weights back to the CosSqrClamp parameter transformation. This proves superior in edge cases.

Other
-----
- Fixed spelling mistake in GUI (misspelled Brain as Brian).

====================

Other
-----
- Bug fix in the create_covariance_matrix, it sometimes tried to get the shape attribute of a dictionary, crashing the computations. This only happened in rare occasions.