NARXNET training with independent data sets

Hi! I'm currently working on a NARXNET project whereby I would like to train the network with a 20 timestep data set which contains the financial ratios of company A and the target output as company A's bond yields. I'd like to know if it's possible/feasible to train this already trained network with the datasets of company B, C and so forth. Thanks in advance!

Answer 1

[Greg Heath]

If you first train with dataset A you will get weight and bias vector

 WA = getwb(netA).

If you then train netA with dataset B, the properties of datasetA stored in WA will be OVERWRITTEN AND FORGOTTEN.

Therefore, when you introduce datasetB, it should be as a subset of the combined datasetAB which contains both datasets.

Hope this helps.

Thank you for formally accepting my answer

Greg