.. _changes:

# Changelog

This document lists changes to ``ionics_fits``.

Changes are divided into four categories:
* Bugfixes
* API changes - changes to the user-facing API. These will often be **breaking** (not backwards-compatible), requiring changes to code using ``ionics_fits``.
* Other - small changes, such as tweaks to internals, which most users will not need to be aware of
* New models

## 1.3.0.11

### Bug fixes
* Fix ``Lorentzian`` model.

## 1.3.0.9

### Bug fixes
* Fix calling ``func`` on ``RabiFlop``

## 1.3.0.8

Fitting to datasets with an arbitrary number x-axis dimensions is now supported!

### API changes
* **New feature**: Fitters now support datasets with multiple x-axis dimensions
* **Breaking change**: scale function definitions have been moved from ``fits.common``
  to ``fits.utils``
* **Breaking change**: "y channels" have been renamed to "y axes"
* **Breaking change**: models must now provide a ``get_num_x_axes`` method
* **Breaking change**: we now use lists of x and y axis scale factors for the various
  dimensions instead of a single boolean to cover all dimensions in each axis
* **Breaking change**: "containers" have been renamed "transformations"
* **Breaking change**: we no longer export classes inside ``_init__.py`` files

## 1.3.0.5

### API changes
* **New feature**: Added ``ReparametrizedModel`` container
* **New feature**: Added ``SineMinMax`` model
* **New feature**: Added ``Sine2`` model
* **Breaking change**: ``rescale_model_x`` has been replaced with `ScaledModel`` to make
  this feature consistent with the rest of the toolkit.

### API changes
* **New feature**: Added a sigmoid model

## 1.3.00

This release includes some invasive refactoring of ``ionics_fits`` internals to simplify
dataflow. This requires small changes to how models are defined (see breaking changes)
but will otherwise be transparent to users.

A major feature addition in this release is ``ionics_fits`` first foray into the world of multi-dimensional fitting.

### Bugfixes
* Incorrect calculation of pi times in the Rabi model

### API changes

* **Breaking**: ``estimate_parameters`` no longer takes a dictionary of model parameters as an input. Instead, it acts directly on the model's parameter dictionary.
* **Breaking**: all model parameters must now specify a scale function. This changes the previous behaviour where parameters with no scale function specified were treated as invariant.
* **Breaking**: scale functions are no longer passed the model as an input argument
* **New feature**: helper functions were introduced to ``ionics_fits.common`` to make specifying scale functions easier and more readable.
* **Breaking**: whether or not a model can be rescaled is now determined by the model's ``can_rescale`` method. All models must provide this method. Scale functions must now always return a float value. This replaces the previous behaviour where scale functions returning ``None`` was used to indicate a model that cannot be rescaled.
* **New feature**: enabling rescaling can now be set independently for the x and y axes
* **Breaking**: ``Model``'s ``get_initial_values`` method has been removed
* **New feature**: ``ModelParameter`` s now raise exceptions if any fixed parameters have user 
estimates. This behaviour was previously allowed but was generally indicative of a bug
* **Breaking**: heuristics have now been moved into ``ionics_fits.models.heuristics``. This includes functions, such as ``param_min_srs``, which were previously methods of ``Model``
* **Breaking**: removed a sampling heuristic which wasn't particularly useful
* **New feature**: solver arguments are now exposed via fitters
* **Breaking**: ``AggregateModel`` now takes a dictionary of models rather than a list of
  tuples.
* **Breaking**: ``AggregateModel`` now names parameters ``{param_name}_{model_name}``
  instead of ``{model_name}_{param_name}``

### New models
* ``ConeSegment`` model has been added

### Other changes
* Added a change log!
* Models now have ``internal_parameters`` as well as ``parameters``. Internal parameters represent parameters which are not exposed directly to the user, but which still need to be stored and rescaled. They are used, for example, in the container models.
* Improved documentation
* The way rescaling is handled has been overhauled to improve dataflow. The ``get_scaled_model`` mechanism has been removed in favour of using ``internal_parameters``. ``ModelParameter``s and ``Model``s now provide ``rescale`` and ``unscale`` methods to control rescaling. When a ``ModelParameter`` is rescaled, querying its attributes returns scaled values.
* ``qt`` is now a development dependency
* replaced ``np.power`` with the more pythonic (and faster) ``**`` operator where suitable
* ownership / copying has been tidied up and documented better
* ``LaserRabi``: improved heuristic for estimating ``eta`` when ``omega`` is known
* Improved parameter estimator for ``RepeatedModel``
* `RepeatedModel` now has the ability to aggregate derived parameter values, reporting
  only the values that are the same for all repetitions.


## Previous versions

Changes were not tracked prior to ``v1.3``.