Jenn

Build

- Made `matplotlib` required dependency (made dev easier to manage)

Fix

- Modified exposed utils

Feat

- Add support for loading JMP models into Python using JENN

Fix

- Change default activation in `Parameters` class from `relu` to `tanh`
- Fix initialization of `sigma_x` and `sigma_y` to use `np.ones` (erroneously, it previously used `np.eye`)

Docs

- Deleted `theory.pdf` (no longer needed now that paper is on ArXiv)
- Updated CONTRIBUTING to reflect `pixi` process (more simple)
- Added section about loading JMP models into JENN (with examples)

Build

- Switched from `doit` to `pixi` (no need for a base environment anymore, more simple overall)
- Update GitHub Actions workflow to use `pixi`

Test

- Added `nbmake` to test example notebooks during `qa`
- Added unit tests for new JMP feature

Docs

- Added link to technical paper on ArXiv (preprint) in README and `docs\index.rst`
- Fixed notation inconsistency in Jacobian matrix (data structures section)
- Updated `demo_4_rosenbrock.ipynb` with plot annotations (and fixed random seed)

Refactor

- Switched order of indices `r` and `s` in `propagation.py` to match paper

Fix

- missing dependencies (`jsonschema`, `jsonpointer`)
- missing data (*.json was not being included in build, so added MANIFEST.in)
- typing oversight for python 3.8 (in `cost.py` and `sythetic.py`)

Fix

- Fixed random seed not working (previously not being passed to parameter initialization)
- Fixed `minibatch` issue throwing error below when `shuffle=False` and more than one batch

Traceback (most recent call last):
File "C:\[...]\jenn\model.py", line 141, in fit
self.history = train_model(
File "C:\[...]\jenn\core\training.py", line 121, in train_model
batches = data.mini_batches(batch_size, shuffle, random_state)
File "C:\[...]\jenn\core\data.py", line 229, in mini_batches
batches = mini_batches(X, batch_size, shuffle, random_state)
File "C:\[...]\jenn\core\data.py", line 51, in mini_batches
if mini_batch:
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()


Refactor

- Added jsonschema to validate reloaded parameters and check array shapes
- Added levels as input to `plot.contours`

Features

- Added optional ability to prioritize individual training points (useful to ensure more accuracy in known regions of interest)
- Added optional ability to warmstart; i.e. continue training from current parameters (without initialization)
- Exposed more hyperparameters pertaining to optimizer (e.g. tolerance stopping criteria)
- Added option to use finite difference for generating synthetic data partials (used to study effect noisy partials)

Documentation

- Added airfoil notebook as example of large dataset
- Added surrogate-based optimization notebook to demonstrate benefit of JENN
- Updated theory.pdf

Fix

- Updated annotations in `jenn.utils.plot` which were incompatible with Python 3.8 (causing runtime errors)
- Manually updated `__version__` number inside `__init__` (previous oversight)

Documentation

- Update demo examples to use `from jenn.utils import plot` instead of `jenn.utils.plot` (which failed a test on Python 3.11.7)