Re: Neural net & walk forward testing.

[mgj@xxxxxxxxxxxx]

From: Kozo Morimoto <K.Morimoto@xxxxxxxxx>
>>  I just don't have enough data to train AND test.<<

This is the most common problem with training models with large degrees of 
freedom: not having enough data to contrain the model.  My recommendation here 
is to randomly select 70% of the original data for training, 30% for testing 
and keep your model AS SIMPLE AS POSSIBLE.    Use the smallest net that 
delivers reasonable performance.   Squeeze all the redundancy you can out of 
the data, thereby reducing the number of input variables for the NN model.

For help on particulars, email to me directly.

- Mark Jurik