In the last post, I added a link to a downloadable spreadsheet where one can play with the ENSO solver.

Interesting what it can do with very short training intervals. This is one from 1908 to 1920, a short training interval of only 12 years. You would think that it would overfit, yet the extrapolated out-of-band model looks reasonable and not at all chaotic

![train](http://imageshack.com/a/img924/6987/URH1w1.png)