There are 840 data files in this archive (plus this README).

The data consist of true acceptor sites (ip_acc_*+1.{val,train,test} =210 files),
true donor sites (ip_don_*+1.{val,train,test} =210 files) and 
corresponding decoys (ip_{acc,don}*-1.{val,train,test} =420 files). The decoys 
for the acceptor consist of acc. decoys and true donor sites.
The decoys for the donor consist of donor decoys and true acceptor
sites.
The data is split up in 10 runs (0-9). For each run there
are 10 splits (0-9) used for 10-fold cross validation (the *.val and *.train
data), like ip_{acc,don}_run[0-9]_split[0-9]_{+,-}1.{train,val} where
+1 stands for the true splice sites and -1 for the decoys as described
above.
Testing was done for each run on the ip_{acc,don}_run[0-9]_{+,-}1.test
files. These results were averaged and
mean and standard deviation were mentioned in the results.
