This is a general purpose engine for an Echo State Network, a type of recursive nerual network. The advantages of this network is the smaller number of weights to train, allowing for faster learning and active example understanding for larger deep networks. In addition, this engine was build to support oscillatory computing, a byproduct of a Hopfield architecture and intelligent network generating.
This repo was created to work similarly to scikit-learn, which is based on training, fitting, and predicting. Once the data has been imported and massaged into shape, the output weights need to be trained or quietly fit so that the ESN can predict more and more accurate outputs. That's about it, you train or fit to set the model to the data so you can predict the correct outputs for different inputs.
Creating an ESN:
myEsn = esn(50, 5, 5)
Fitting:
myEsn.fit(train_x, train_y)
Predict over single input (overdampened):
myEsn.predict(my_input, 0)
Scoring:
myEsn.score(test_x, test_y)
ESN Engine v0.9