v0.4.0

What's Changed

• Ensure that the PSDEstimator inherits from the BaseEstimator by @kevincar in #47
• 48 windowing optimizations by @kevincar in #49
• Add float 64 to csp computation by @kevincar in #50
• Develop by @kevincar in #51
• 52 check fit params by @kevincar in #53
• Update the pwr function to decibels by @kevincar in #55
• Develop by @kevincar in #56

Full Changelog: v0.3.12...v0.4.0

v0.3.9

Update to transformers

v0.3.12

Add preprocessing of np.ma.MaskedArray objects so that they are processable and cashable through joblib memory functions

v0.3.1

Added

• n-dimensional standard scaler
• n-dimensional min max scaler

Changed

• Updated n-dimensional filtering on the -3 axis for filter bank

v0.3.0

Changed

• NDTemporalFilter for multidimensional temporal filtering
• General masked classes (e.g., PSDBinner) now simply pass the data to the non-mask version if there is no specific mask applied to the data
• TransformPipeline now has predict
• Transformers (e.g., Windower and Labeler) now only do processing during the transform step to ensure appropriate processing and avoid the need to fit when fitting is not semantically correct to do.
• Windower can not return masked data if label_scheme is 3 or 4.

Added

• Decorrelate class added for pruning correlated feature vectors
• Common features
• Mask-friendly metrics: accuracy_score
• Multidimensional and mask-friendly CSP for general and filter-bank CSP
• OptionalTransformer for allowing pipelines to completely ignore a step of the pipeline
• Utility functions

v0.2.1

Changes

• TransformPipeline now tracks results when calling transform or fit_transform. Addressed in issue #14
• TransformPipeline also uses appropriate transform and fit semantics
• Windower and Labeler should not use fit, only transform. This is address with issue #15

v0.2.0

Changed

• set_params added to many of the mne classes to ensure pipelines would be compatible with grid search.
• Labeler now returns a 3D array in the shape of (n, channels, time) where n is the number of files given on input. time is the time length of each file, filter across labels provided
• ‘Windower’ is not agnostic to the time dimension and can be specified to split the time dimension in to windows and window lengths. When coupled with the Labeler this now outputs the following shape (n, channels, windows, time). For this to work with standard pipelines a second transformation must occur to merge the windows and n appropriately.