Pycmtensor

Feat

- **optimizers.py**: include new optimizer RProp
- **functions.py**: speed up computation and compilation by using static indexing in `log_likelihood` function
- **functions.py**: add `relu()` function (source taken from Theano 0.7.1)
- **basic.py**: new function include_params_for_convergence
- **dataset.py**: use `as_tensor_variable` to construct tensor vector from dataset[[item1, item2,...]]
- **dataset.py**: added list(`str`) as tensor arguments for `train_dataset()` and `valid_dataset()`

Fix

- **__init__.py**: fix init circular imports
- **statistics.py**: update t_test and varcovar matrix calculations for vector parameters
- **layers.py**: various fixes to neural net layers
- **optimizers.py**: fix SQNBFGS algorithm
- **functions.py**: include output as params for 1st and 2nd order derivatives
- **expressions.py**: fix base class inheritence and use `config.seed` for seed value
- **expressions.py**: include `self` in overloaded operators if `other` instance is of similar type as `self`
- **basic.py**: fix incorrect saved params in `train()`
- **expressions.py**: clip lower and upper bounds when updating `Betas`
- **expressions.py**: fixed `Weights` and `Bias` mathematical operations
- **config.py**: renamed config.py to defaultconfig.py to avoid name conflicts

Refactor

- **results.py**: updates results calculations for beta vectors
- **expressions.py**: add set_value in `Param` base class
- **optimizers.py**: removed unused imports

Feat

- **optimizers.py**: added new BFGS algorithm
- **dataset.py**: moved dataset initialization to dataset.py
- **pycmtensor.py**: Implemented early stopping on coefficient convergence in training loop
- **functions.py**: logit method now takes uneven dimensioned-utilities
- **expression.py**: Added RandomDraws expression for sampling in mixed logit
- get_train_data optional argument numpy_out to return numpy arrays rather than pandas arrays
- BHHH algorithm for calculating var-covar matrix applies to each data row

Fix

- **results.py**: fixed instance when params are not Betas
- **expressions.py**: include RandomDraws as an option in expression evaluations
- Update tests workflow file conda packages
- **pycmtensor.py**: Added missing configuration for stepLR `drop_every`
- **tests.yml**: Update tests workflow file conda packages
- **optimizers.py**: Fixed name typo in `__all__`
- **results.py**: Corrected calculation of hessian and bhhh matrices
- **scheduler.py**: Moved class function calls to parent class
- **statistics.py**: Fixed rob varcovar calculation error
- **MNL.py**: Moved aesara function to parent class
- **data.py**: Streamlined class function calls and removed unnecessary code
- removed package import clashes with config.py
- removed gnorm calculation
- update hessian matrix and bhhh algorithm functions

Refactor

- **pycmtensor.py**: temporarily removed pycmtensor.py
- **MNL.py**: replaced function constructors from pycmtensor.py as a function call inside the model Class object
- **basic.py**: moved model functionality to from pycmtensor.py to models/basic.py
- **utils.py**: Removed unused code

Fix

- make arguments in `MNL` as optional keyword arguments
- moved learning rate variable to `PyCMTensorModel` class

Refactor

- make model variables as property
- update `__all__` package variables
- added `train_data` and `valid_data` property to `Data` class

Fix

- fix utility dimensions for asc only cases

Feat

- **optimizers**: added ``Nadam`` optimizer
- **layers.py**: added ``DenseLayer`` ``BatchNormLayer`` ``ResidualLayer``
- added ``pycmtensor.about()`` to output package metadata
- added EMA function ``functions.exp_mov_average()``

Fix

- renamed depreceated instances of ``aesara`` modules
- **data.py**: defaults ``batch_size`` argument to 0 if batch_size is ``None``
- updated syntax for ``expressions.py`` class objects
- added ``init_type`` property to ``Weights`` class
- moved model aesara compile functions from ``models.MNL`` to ``pycmtensor.PyCMTensorModel``
- added argument type hints in function.py

Refactor

- **data**: added import dataset cleaning step as arguments in `Data()`
- moved ResidualLayer to ``pycmtensor.models.layers``
- updated timing to perf_counter
- **pycmtensor**: refactoring model_loglikelihood

Feat

- added ``pycmtensor.about()`` to output package metadata
- added EMA function ``functions.exp_mov_average()``

Fix

- updated syntax for ``expressions.py`` class objects
- added ``init_type`` property to ``Weights`` class
- moved model aesara compile functions from ``models.MNL`` to ``pycmtensor.PyCMTensorModel``