Enable cross-validation and mini-batch options for xfiles-dana. This generally involves a rewrite of the changes available in a6eb0b1.
With this you should be able to specify a training/validation or training/validation/testing split and then report the MSE for each. Cross validation should not be enabled by default, but it may make sense to use cross validation in the 01-fann-xfiles regression test.
@ldelshad, this should be much easier to handle with the refactored fann-xfiles (c979e49) which removes the code replication and defines generic incremental and batch training functions xfiles_train_batch and xfiles_train_incremental. Both of these will train on a specific range of items in the data set. I expect that these can be directly used to operate on subsets of the data for cross-validation and mini-batches.
Enable cross-validation and mini-batch options for xfiles-dana. This generally involves a rewrite of the changes available in a6eb0b1.
With this you should be able to specify a training/validation or training/validation/testing split and then report the MSE for each. Cross validation should not be enabled by default, but it may make sense to use cross validation in the 01-fann-xfiles regression test.
@ldelshad, this should be much easier to handle with the refactored
fann-xfiles
(c979e49) which removes the code replication and defines generic incremental and batch training functionsxfiles_train_batch
andxfiles_train_incremental
. Both of these will train on a specific range of items in the data set. I expect that these can be directly used to operate on subsets of the data for cross-validation and mini-batches.