cknd
commented
6 years ago Hi! Right now, that probably won't work since it's all implemented very naively: The fit method always starts from a 0 state, then fills a buffer of states the size of the given input and then overwrites the weights by the best fit on only the states produced in this call to fit. For it to work with small batches (down to N=1) over subsequent calls, it would need to remember the network state between calls to fit and use online weight updates. Do you feel up to the task of developing this?
Theoretically, ESN model could be trained incrementally online. However, It seems that pyESN model does not behavior as expected when I tried to fit in two bathes as compared to fit in a whole batch. Could you please tell me if pyESN model support fitting incrementally or not? Thanks for your excellent work.