Mdt

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

Other
-----
- Removed atomic function from the KurtosisTensor, it was redundant.

====================
This version is the first version to support arbitrary non-linear inequality constraints.
Currently, during optimization, these constraints are enforced using the penalty method (https://en.wikipedia.org/wiki/Penalty_method).

Added
-----
- Adds special parameter classes for the spherical parameters. Now, the composite model will take care of the necessary transformations to keep the spherical parameters theta and phi within the right spherical hemisphere.
- Adds support for inequality constraints to the compartments and composite models.
- Adds utility function for merging dictionaries in inheriting component templates.

Changed
-------
- Changed parameter transformation of the weights from CosSqrClamp to ScaleTransform

Removed
-------
- Removed the post-optimization modifiers. They were unnecessary since capability of specifying constraints.

Other
-----
- Added files for docker/singularity intel builds

====================
- Corrected MOT version requirements.

====================
The primary change in this version is that we now use the pseudo-inverse for computing the covariance matrix from the Hessian.
This is as good as a direct inverse but faster to compute and more robust to voxels with a badly conditioned Hessian.

Added
-----
- Adds support for masked colors setting in the maps visualizer.

Changed
-------
- Changed Fisher Information Matrix computation to use an eigen-decomposition based pseudo-inverse for all voxels.
- Changed the signature of the batch_apply function to have the path as first argument. The subject info of the batch functions now include the data folder as property.

Fixed
-----
- Fixed small issue in the GUI where shells were not correctly counted and represented due to a sorting issue in the protocol class.

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

Changed
-------
- Improved the Hessian computation of the models after fitting.
- Changed the default BinghamNODDI_r1 initialization to depend on the Watson NODDI model.

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

Changed
-------
- Adds model_names option to the run_function_on_batch_fit_output batch function.
- Removed some numpy warnings in the qMT model.
- Increased the JohnsonNoise eta upper bound.