This is the changelog for the rescomp package
- Add Scipy 1.8 support by removing a depricated import
- Fixed ESNGenLoc bug where W_in was setup incorrectly.
- Updated the examples to use the Lorenz-96 system instead of the Kuramoto-Sivashinsky system where applicable.
- Removed the measures.error_over_time() and measures.valid_time_index() functions introduced in 0.2.4 due to concerns about their naming. They'll be re-introduced in a later version.
- Added measures.error_over_time() and measures.valid_time_index() as additional error/data validation functions
- Added esn.create_input_matrix() to allow the creation of an input matrix independent from any training.
- Adjusted the arguments of _ESNCore._fit_w_out() to allow for more flexible extension classes built upon _ESNCore and/or ESN
- Updated required packages to correctly reflect the oldest supported packages.
- Removed the last matplotlib uses in rescomp.
- Added better testing with tox (and pyenv)
- Added tests for the KS simulation function variants, and marked the (likely) expected to fail tests as such.
-
Fixed Bug where the parameter pred_sync_steps was not passed on and used incorrectly in the following functions:
- esn.ESNWrapper.train_and_predict()
- esn.ESNGenLoc.train_and_predict()
- utilities.train_and_predict_input_setup()
Directly called predict() functions, the typical use case where an additional synchronization is required, are unaffected.
- Updated the repositories license from unlicensed to the MIT license
- Added the general local states (GLS) Code
- Added a usage example for the general local states methods
- Rewrote the documentation for gitlab pages
- Rewrote the examples as jupyter files for integration in the new documenation
- Added options to have W_in ordered
- Implemented squared tanh activation function
- Implemented option to have mixture of activation functions
- Changed the 'b' parameter in the lorenz systems used in rescomp.simulate_trajectory() to 'beta', as that's what the parameter is usually called in the literature
- Allowed kwargs for the Lorenz63 and Roessler systems in simulate_trajectory()
- Added short example for the simulations module usage.
- Rewrote divergence_time to correctly handle perfect predictions and immediate divergence
- Package fundamentally rewritten.
See README, HTML documentation and examples for details.