Pycmtensor

Feat

- **utils**: add new function to display numbers as human readable format
- **results**: add new function to plot training statistics

Fix

- **scheduler**: fix missing arguments
- **basic**: renamed variable to avoid confusion

Fix

- **scheduler**: fixed calculation erro in cyclic lr
- condition for stdout

Feat

- new option for selection of model acceptance pattern
- calculating distributed Beta now uses percentile instead of mean
- **models**: add new func `compute` to run the model on a manual set of params
- **regularizers**: new method `include_regularization_terms()` in BaseModel for adding regularizers

Fix

- **dependencies**: prevent breakage of numpy 1.26
- import compute from pycmtensor.models
- fix l2 regularizer formula
- set init_types as a module level list and add tests
- add minmax clipping on neural net layer outputs
- temporary function for negative relu
- expression fix for random draw variables

Refactor

- relax the condition for stdout during training
- **config**: refactor config.add into .add() and .update() methods

Refactor

- changed basic model functions into staticmethod
- **expressions.py**: make ExpressionParser.parse a staticmethod
- remove depreciated modules

Fix

- **expressions.py**: use predefined draw types for random draws
- **basic.py**: include choice label(s) as a dictionary for `predict()`
- moved elasticities from statistics to model

Fix

- **scheduler.py**: add learning rate lower bounds for decaying functions
- **basic.py**: add placeholder arguments `*args`

Refactor

- **basic.py**: improve efficency of hessian matrix calculation over the sum of log likelihood over observation
- **basic.py**: refactoring common model functions into BaseModel
- syntax and naming changes