How about interleaving the test data then, instead of appending it to the very end? For gzip, if the block size is 64k (another comment says 32k?), split the co

How about interleaving the test data then, instead of appending it to the very end? For gzip, if the block size is 64k (another comment says 32k?), split the corpus text into 32k blocks, and interleave it with 32k blocks of the test set.