echos 1.0.4
Improvements
- Expanded package documentation, vignettes and README to better
describe the ESN architecture, key hyperparameters, model selection
procedure and tuning workflow.
- Clarified the role of the
formula argument in
ESN() for the fabletools model interface.
Datasets
- Renamed
m4_data to m4_monthly_subset to
clarify that the dataset contains a monthly subset of the M4 competition
data. The old name m4_data is retained for backward
compatibility.
- Improved documentation for
m4_monthly_subset and
synthetic_data.
echos 1.0.3
New features
- Added reservoir scaling parameter
tau to
train_esn(), enabling dynamic control of reservoir
size.
- Added
tune_esn() to tune hyperparameters
alpha, rho and tau via time
series cross-validation (i.e., rolling forecasts).
- Added S3 methods
summary.tune_esn() and
plot.tune_esn() to summarize and visualize results from
hyperparameter tuning.
Bug fixes
- Fixed
train_esn() so n_initial is only
auto-set when NULL.
Improvements
- Added input validation for
y and inf_crit
in train_esn() and levels in
forecast_esn().
- Improved documentation
echos 1.0.2
New features
- Added forecast intervals to
forecast_esn(),
forecast.ESN() and plot.forecast_esn().
Forecast intervals are generated by simulating future sample path based
on a moving block bootstrap of the residuals and estimating the
quantiles from the simulations.
- Added
plot.esn() to visualize the internal states
(i.e., the reservoir).
- Added
filter_esn() to extract ESN models from a
mable.
- Added
synthetic_data, a dataset with synthetic time
series data as tibble.
Improvements
- Improved documentation
- Added unit tests
- Reduced dependencies
echos 1.0.1
- Updates due to CRAN comments
echos 1.0.0