small improvement by training with part of data converted to lower case

first release