Releases · aims-umich/pyMAISE

What's Changed

New neural network structure introduced in #48, which includes support for the following layers:
- dense,
- dropout,
- LSTM,
- GRU,
- 1D, 2D, and 3D convolutional,
- 1D, 2D, and 3D max pooling,
- reshape,
- and flatten.
Includes support for the following optimizers:
- Adam,
- SGD,
- RMSprop,
- AdamW,
- Adadelta,
- Adagrad,
- Nadam,
- and Ftrl.
Added hyperparameter classes for neural networks:
- pyMAISE.Int
- pyMAISE.Float
- pyMAISE.Boolean,
- pyMAISE.Choice,
- and pyMAISE.Fixed.
Added tuners for neural networks:
- grid,
- random,
- Bayesian,
- and hyperband search.
Classification models implemented in #44 , #48:
- logistic regression,
- decision tree,
- k-nearest neighbors,
- random forest,
- support vector machine,
Additional functions and metrics for classification:
- confusion matrix
- metrics including precision, recall, F1, and accuracy,
- and one-hot encoding.
Converted data structures for raw data to xarray.DataArrays
All load functions now return xarray.DataArrays
Preprocessing is now a module of functions and classes
Added pyMAISE.preprocessing.SplitSequence for handling 2D and 3D time series data
Added pyMAISE.datasets.load_loca() for LOCA data, which will be incorporated into a future benchmark
Updated documentation to reflect changes

Added CI testing that incorporates automatic testing using Python 3.9, 3.10, and 3.11
Added pre-commit hook for enforcement of PEP 8 standards and formatting

Full Changelog: https://github.com/myerspat/pyMAISE/commits/1.0.0b0