The echo state network (ESN) is a special recurrent neural network model proposed by H. Jaeger in 2001, which is a representative model for reservoir computing. The diverse-timescale ESN (DTS-ESN) (Tanaka et al., Phys. Rev. Res. 2022) is an extended ESN model with a rich variety of timescales for prediction of multiscale dynamics. The python codes for the model and demonstrations are provided.
- 1step_ahead_prediction: this folder contains a sample code and a demo for reproducing Fig. S2(e) in the supplementary material of the reference paper.
- longterm_prediction: this folder contains a sample code and a demo for reproducing Fig. 5(b) of the reference paper.
Some python modules, such as numpy, scipy, matplotlib, and networkx, are required to run the codes.
Gouhei Tanaka, International Research Center for Neurointelligence (IRCN), The University of Tokyo
G. Tanaka, T. Matsumori, H. Yoshida, K. Aihara, "Reservoir computing with diverse timescales for prediction of multiscale dynamics," Physical Review Research, vol.4, L032014 (2022)