Spacekit

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

installation
------------
- Set minimum python version to 3.10, include py312 in test suite, replace deprecated pkg_resources with importlib.metadata [73]
- Set minimum tensorflow version to 2.16.1 [73]
- Pre-trained neural networks updated for compatibility with Keras 3 [73]
- Dockerfile images now use debian bookworm [73]

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

new features
------------
- `builder.trained_networks` jwst_cal.zip includes updated (v2) `img3_reg` and new `spec3_reg` predictive models for image and spectroscopic data [58]
- `preprocessor.ingest.JwstCalIngest` class and cmdline script for automated training data ingest [57]
- `extractor.radio.JwstCalRadio` subclass for scraping datasets from MAST using ASN metadata [51]
- `extractor.scrape.FitsScraper.scrape_dataframe` method added for scraping Fits data from dataframe [52]

enhancements
------------

- `skopes.jwst.cal.predict` generates predictions for spectrosopic datasets in addition to image data. This update also allows further customization of user arguments: [58]
- `obs` to specify selection of a program ID + observation number
- `input_path` accepts either a directory (default) or a filename. If filename, the script will try to find any input exposures that belong to the same program and observation number as that file.
- `sfx` attribute is now customizable on instantiation of the class object (default is '_uncal.fits')
- `architect.builder.Builder.save_model` uses preferred keras archive format by default [50]
- `preprocessor.transform.SkyTransformer` set offsets to 0 for gs/targ fiducial NaN values; custom filename for tx_file [54]
- `preprocessor.prep.JwstCalPrep` updates in preparation for preprocessing spectroscopic data [55]
- revise spectroscopic data columns
- save tx_file name with "-{expmode}" to differentiate between image and spec normalization params
- rename target attributes: y_img_train, y_img_test to y_reg_train, y_reg_test
- `preprocessor.scrub.JwstCalScrubber` more sophisticated exposure grouping and L3 product naming [56]

bug fixes
---------
- `preprocessor.encode.PairEncoder.handle_unknowns` create single new encoding value per unidentified variable [53]

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

bugfixes
--------

- move HstSvmRadio import inside class method to avoid importing astroquery unnecessarily [49]

- temporarily pin tf max version to 2.15 to ensure compatibility with models saved in 2.13 or older

- matplotlib style setting looks for "seaborn-v0_8-bright" if "seaborn-bright" unavailable, fallback uses default style


installation / automation
-------------------------

- GA workflow minor revision: pypi publish [46]

- Replace flake8 with ruff, replace deprecated tf.keras.wrappers.scikit_learn with scikeras, add GA workflows [45]

documentation
-------------

- Update readthedocs.yaml for compatibility with latest formatting requirements [44]

- RTD: Install graphviz before building docs [47]

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

- New feature: JWST Calibration Processing resource prediction model and skope (prediction script) added under the architecture name "jwst_cal"

- Pretrained neural network files and paths renamed: calmodels.zip is now hst_cal.zip,
ensemble.zip is now svm_align.zip.

- If keras models are saved using the older SavedModel format, you must pass `keras_archive=False` when loading a saved model. By default, new models will be saved using the newer keras archive format.

- Tests added for JWST; existing tests and metadata updated to reflect above changes

- Updated zenodo version ID for remote test data

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

- bugfix set dataframe columns with bracket instead of curly bracket (resolves pandas>1.4 incompatibility)

- remove pandas pinned version

- improved log handling with spacekit/logger module

- added predict script for hst cal skope

- updated docker dashboard templates

- enhancements for loading pretrained models

- pytest configuration updates and new tests added

- plugin for external test data

- updated repo url badges

- updated documentation

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

- bugfix scikit-learn replaces deprecated sklearn dependency

- temporarily pinned `pandas` dependency to 1.4.x and below due to column setting bug in v1.5

- bugfix keras `load_img` method imported from tf.keras.preprocessing.image instead of tf.keras.utils

- new feature skopes.hst.cal model training, inference, cross-validation scripts added

- new feature svm dashboard predict view

- svm ensemble model archive file `ensembleSVM.zip` renamed as `ensemble.zip`. This extracts to `models/ensemble/` with `tx_data.json` (transform data) and `ensembleSVM` (keras model binaries) inside of the `ensemble/` parent directory. Previously, the json file was inside ensembleSVM alongside the binaries.